do eu structural funds promote regional employment? evidence from dynamic … · 2011-12-13 ·...
TRANSCRIPT
WORK ING PAPER SER I E SNO 1403 / DECEMBER 2011
by Philipp Mohland Tobias Hagen
DO EU STRUCTURAL FUNDS PROMOTE REGIONAL EMPLOYMENT?
EVIDENCE FROM DYNAMIC PANEL DATA MODELS
1 European Central Bank, Kaiserstr. 29, 60311 Frankfurt, Germany, e-mail: [email protected]
2 Frankfurt University of Applied Sciences, Nibelungenplatz 1, D-60318 Frankfurt am Main, Germany;
e-mail: [email protected]
This paper can be downloaded without charge from http://www.ecb.europa.eu or from the Social Science Research Network electronic library at http://ssrn.com/abstract_id=1963699.
NOTE: This Working Paper should not be reported as representing the views of the European Central Bank (ECB). The views expressed are those of the authors
and do not necessarily reflect those of the ECB.
WORKING PAPER SER IESNO 1403 / DECEMBER 2011
DO EU STRUCTURAL
FUNDS PROMOTE
REGIONAL EMPLOYMENT?
EVIDENCE FROM DYNAMIC
PANEL DATA MODELS
by Philipp Mohl 1 and Tobias Hagen 2
In 2011 all ECBpublications
feature a motiftaken from
the €100 banknote.
© European Central Bank, 2011
AddressKaiserstrasse 2960311 Frankfurt am Main, Germany
Postal addressPostfach 16 03 1960066 Frankfurt am Main, Germany
Telephone+49 69 1344 0
Internethttp://www.ecb.europa.eu
Fax+49 69 1344 6000
All rights reserved.
Any reproduction, publication and reprint in the form of a different publication, whether printed or produced electronically, in whole or in part, is permitted only with the explicit written authorisation of the ECB or the authors.
Information on all of the papers published in the ECB Working Paper Series can be found on the ECB’s website, http://www.ecb.europa.eu/pub/scientific/wps/date/html/index.en.html
ISSN 1725-2806 (online)
3ECB
Working Paper Series No 1403December 2011
Abstract 4
Non-technical summary 5
1 Introduction 7
2 Econometricspecification 102.1 Baseline panel approach 102.2 Spatial panel approach 152.3 Panel approach with interaction term 17
3 Econometric results 183.1 Baseline panel approach 183.2 Spatial panel approach 223.3 Panel approach with interaction term 23
Conclusions 25
Acknowledgements 27
28
Tablesandfigures 30
References 43
CONTENTS
Appendix
4ECBWorking Paper Series No 1403December 2011
Abstract
Despite its rather broad goal of promoting “economic, social and terri-torial cohesion”, the existing literature has mainly focused on investi-gating the Cohesion Policy’s growth effects. This ignores the fact thatpart of the EU expenditures is directly aimed at reducing disparitiesin the employment sector. Against this background, the paper ana-lyses the impact of EU structural funds on employment drawing ona panel dataset of 130 European NUTS regions over the time period1999-2007. Compared to previous studies we (i) explicitly take intoaccount the unambiguous theoretical propositions by testing the con-ditional impact of structural funds on the educational attainment ofthe regional labour supply, (ii) use more precise measures of structuralfunds for an extended time horizon and (iii) examine the robustnessof our results by comparing different dynamic panel econometric ap-proaches to control for heteroscedasticity, serial and spatial correlationas well as for endogeneity. Our results indicate that high-skilled pop-ulation in particular benefits from EU structural funds.
Keywords: EU structural funds, dynamic panel models, spatial paneleconometrics, regional employment effectsJEL classification: R11, R12, C23, J20
5ECB
Working Paper Series No 1403December 2011
Non-technical summary
The largest part of the EU budget–more than one third of total EU ex-
penditures and corresponding to 380 billion euros–is spent on EU Cohesion
Policy for the Multiannual Financial Framework from 2007–2013. Despite its
rather general focus on promoting “economic, social, and territorial, cohesion
among Member States” (Art. 3(3) TEU), the investigation of the impact of
Cohesion Policy has mainly concentrated on the policy’s growth effects.
However, the employment effects are key to understanding regional in-
come disparities, since income differences are, per definition, based on dif-
ferences in the labour productivity and/or employment level, among other
factors. In addition, parts of the EU expenditures (in particular Objective 2
payments) are directly aimed at reducing disparities in the employment sec-
tor. Nevertheless, only a few papers have analysed the employment effects of
this policy field and the overall empirical results are inconclusive. Moreover,
from a theoretical perspective, higher expenditures on EU funding do not
necessarily increase the total employment level. Instead, its impact depends
on whether structural funds are used as capital subsidies or as human capital
investment and it is subject to the educational attainment of the population.
Furthermore, to the extent that structural funds payments have short-term
business cycle effects, the employment effect may be low for economies with
a positive output gap and a tight labour market situation. All in all, the net
effect on total employment is an empirical question.
This question is addressed in this paper, analysing the impact of EU
structural funds on employment with a panel dataset of 130 European regions
over the time period 1999-2007. Our empirical results confirm the theoretical
predictions as total structural funds have no significant positive impact on the
regional employment level. However, we find evidence that structural funds
may be interpreted as capital subsidies and are only conditionally effective.
These funds have a significant positive impact on the total employment level
in regions with a low share of low-skilled population, and have a negative
effect in the case of a high share of low-skilled population.
6ECBWorking Paper Series No 1403December 2011
Our results have policy implications for the setup of future Multiannual
Financial Frameworks. It becomes evident that EU funding lacks a clear
concept on how to promote employment in the medium- to long-run. Our
results indicate that the high-skilled population in particular benefits from
EU structural funds payments. As a consequence, a strategy should define
objectives which are clearly measurable and allow for an ex-post assessment
of this policy field. This, in turn, would contribute to a more effective policy.
7ECB
Working Paper Series No 1403December 2011
1 Introduction
The largest part of the EU budget–more than one third of total EU ex-
penditures and corresponding to 380 billion euros–is spent on EU Cohesion
Policy for the Multiannual Financial Framework from 2007–2013. Despite its
rather general focus on promoting “economic, social, and territorial, cohe-
sion among Member States” (Art. 3(3) TEU), the current literature on the
effectiveness of EU funding has mainly focused on the question whether EU
aid has promoted economic growth and convergence (for a survey see Esposti
and Bussoletti, 2008; Hagen and Mohl, 2011b).
However, the employment effects are key to understanding regional in-
come disparities (measured, e.g. as GDP per capita), since income differ-
ences are, per definition, based on differences in the labour productivity
and/or employment level, among other factors. In addition, parts of the
EU expenditures (in particular Objective 2 payments) are directly aimed at
reducing disparities in the employment sector. Nevertheless, only a few pa-
pers have analysed the employment effects of this policy field. While earlier
contributions find positive employment effects from the European Regional
Development Fund for EU regions in the 1988-1993 period (Busch, Lichtblau,
and Schnabel, 1998) and for firms in northern and central Italy (Bondonio
and Greenbaum, 2006), the recent evidence is rather disillusioning; suggesting
that there are no positive employment effects for EU countries (Heinemann,
Mohl, and Osterloh, 2009) or regions (Dall’erba and Le Gallo, 2007; Becker,
Egger, and von Ehrlich, 2010). By contrast, Bouvet (2005) finds a positive ef-
fect of EU aid on employment growth in a sample of eight EU Member States
between 1975 and 1999.1 One drawback in the literature is the poor data
availability of EU funding. The annual reports on structural funds published
by the European Commission (1995, 1996a,b, 1997, 1998, 1999, 2000) only
1Apart from the studies cited, there is a growing literature which analyses more general
labour market effects at the regional level in Europe, e.g. studies on the determinants
of unemployment (Basile and de Benedicits, 2008) or labour force participation rates
(Elhorst and Zeilstra, 2007).
8ECBWorking Paper Series No 1403December 2011
comprise regional commitments / payments for the 1994–1999 period. Un-
fortunately, since 2000, these reports have only consisted of aggregate data at
the country level. As a consequence, no paper has analysed the employment
effects using regional structural funds payments post 1999.
There are at least four theoretical arguments why EU funding is not un-
ambiguously associated with positive total employment effects. First, struc-
tural funds payments increase the employment level if they lead to human
capital investment (for example, from the European Social Fund); however, if
they are used as capital subsidies (for example, investment grants for firms or
business start-ups), the employment effects will be inconclusive. On the one
hand, structural funds payments reduce capital costs, which leads to more
output and employment (scale effect). On the other hand, reduced capital
costs increase relative costs of labour, which may cause (low-skilled) labour
to be substituted by capital (substitution effect). According to the “capital-
skill-complementary hypothesis” (Griliches, 1969), the demand for skilled
labour increases with decreasing capital costs, while the demand for unskilled
labour decreases with diminishing capital costs. Second, the employment ef-
fects are inconclusive if structural funds payments have a positive effect on
technological progress. According to the “skill-based technological change
hypothesis” (Berman, Bound, and Griliches, 1994), technological progress
may lead to an increase of the relative demand for high-skilled labour, and
thus to a decrease in demand for low-skilled labour. Third, in order to in-
duce a positive employment effect, the regional labor supply must match
with the additional demand for high-skilled labour. Fourth, short-term busi-
ness cycle effects might impede employment growth. If Cohesion Policy was
associated with a positive aggregate demand stimulus and if the economy
was characterised by a positive output gap and a tight labour market sit-
uation, Cohesion Policy would not promote employment growth but would
lead to an overheating of the economy, implying an acceleration of price and
wage inflation. As indicated by Kamps, Leiner-Killinger, and Martin (2009)
this could be in particular the case for the eastern European Member States,
9ECB
Working Paper Series No 1403December 2011
which joined the EU in 2007 and exhibited high growth rates.
All in all, the net effect on total employment is theoretically unclear ex
ante and, hence, an empirical question which is addressed in this paper. We
evaluate the impact of EU structural funds on employment with a panel
dataset of 130 European NUTS regions over the time period of 1999-2007.
Compared with previous studies we explicitly take into account the unam-
biguous theoretical propositions by investigating the conditional impact of
structural funds on the educational attainment. Moreover, we are, to the
best of our knowledge, the first who analyse this research question with more
precise measures of EU funding by distinguishing between Objective 1, 2 and
3 payments and for an extended time period using data from the Multiannual
Financial Framework 2000-2006. Finally, we examine the robustness of our
results by comparing different dynamic panel econometric approaches, high-
lighting specific methodological problems, controlling for heteroscedasticity,
serial and spatial correlation, as well as for endogeneity.
Our results indicate that structural funds in total have no significant
positive impact on the regional employment level. However, we find some
evidence that structural funds are conditionally effective and may be in-
terpreted as capital subsidies. They have a significant positive impact on
the employment level in regions with a low share of low-skilled population,
whereas they have a negative effect on the employment level in the case of
a high share of low-skilled population. This implies that the high-skilled
population in particular benefits from EU structural funds payments.
The outline of this paper is as follows. We start in Section 2 with a
discussion of the econometric specification. This is followed by a presentation
of the empirical results in the light of the methodological challenges in Section
3. Finally, Section 4 concludes.
10ECBWorking Paper Series No 1403December 2011
2 Econometric specification
2.1 Baseline panel approach
Our estimation of the employment effects of structural funds payments is
based on a reduced-form approach including the implications of both a labour
demand model as well as a labour supply model. We define employment
(emp) as the regions’ total employment per population aged 15 to 65 in
order to account for the substantial differences in the size of the regional
labour markets in Europe.
From a theoretical point of view, structural funds payments may affect
employment through the channel of labour demand by increasing the endow-
ment of private and public capital in the region. This raises the marginal
product of labour, the output level, and thus, ceteris paribus, labour de-
mand. A second transmission channel is an increase in the technological
progress which may affect total labour demand positively or negatively, as
discussed in the introduction.
Our baseline specification is defined as follows:2
empi,t = β0 + β1 empi,t−1 + β2 comp.empi,t−1 + β3 pop.youngi,t−1+
+ β4 low skilledi,t−1 + β5 market potentiali,t−1 + β6 grri,t−1+
+ β7 union densityi,t−1 + β8 sfi,t−1 + µi + λt + ui,t
(1)
where the subscript i = 1, ..., 130 denotes the region and t indicates the
time index of our sample for the time period of 1999–2007. Note that all
independent variables are lagged and expressed in log terms. We estimate a
dynamic panel data model by including the lagged employment variable in
order to deal with the sluggish adjustment process (Nickell, 1987).
Moreover, we consider a number of region-specific control variables. We
have to proxy the regional wage level by the compensation of employees in
2Note that we also tested for a non-linear relationship between structural funds and em-
ployment. Our findings, which are available upon request, show that there is no evidence
for a non-linear relation.
11ECB
Working Paper Series No 1403December 2011
the manufacturing sector (comp.emp) due to data availability. Note that the
regional wage level is endogenous with respect to the regional employment
level which has to be taken into account in the estimation strategy (Topel,
1986).
The percentage share of the population aged under 15 (pop.young) is
added as a proxy for two unobserved variables which are relevant for the
quantity and quality of regional labour supply, namely (i) the amount of ex-
perience in the labour market (human capital) and (ii) the effect of having
young children (Elhorst, 2003). We control for the share of population with
low levels of education (low skilled), since the demand for low-skilled work-
ers decreases according to the hypothesis of skill-based technological change
cited above. Hence, an increase in (high-skilled) labour demand may not
raise employment in regions with a high share of low-skilled people, due to
mismatch problems. Moreover, we follow Basile and de Benedicits (2008) for
our definition of market potential. This measure accounts for both the size
of the regional market and its position relative to other regional markets. It
is calculated as the sum of GDP of region i and the weighted GDP of the
neighbouring regions, while the latter is weighted with its squared geograph-
ical distance between the centroids of the countries (the coordinates of the
regional centroids are available upon request).
Furthermore, the scope of the regions in promoting employment is con-
strained by national labour market institutions. As a consequence, we take
into account the level of benefits (Holmlund, 1998) by including the gross
replacement rate (grr). In addition, we control for union density since
higher union density could strengthen the bargaining position of the union,
resulting in higher wage demands and/or a more compressed wage structure
(Scarpetta, 1996; Nickell and Layard, 1999; Blau and Kahn, 1999; Nickell,
Nunziata, and Ochel, 2005).
Moreover, we included the employment protection indicator of the OECD
to account for employment protection laws following the literature by Lazear
(1990). We also considered indicators measuring the structure of the econo-
12ECBWorking Paper Series No 1403December 2011
mies, such as the share of regional employment in the agricultural/industrial
sector. However, the latter variables–the employment protection indicator
and the share of industry structure in regional employment–are not statisti-
cally significant, so we excluded them from our final specifications.
The main variable of interest is the structural funds variable (sf). Table
4 clarifies that total structural funds can be classified into three different ob-
jectives. Around two-thirds of total structural funds payments are allocated
to regions with a GDP lower than 75% of the EU average. This Objective
1 funding has the primary goal to promote development in less prosperous
regions. The remaining part is spent without a clear allocation scheme on
regions in structural decline (Objective 2) and to support education, training
and employment policies (Objective 3). For our empirical analysis we draw
on a dataset, which has, to the best of our knowledge, only been used by
Mohl and Hagen (2010) in the context of the evaluation of economic growth
effects of EU funding. This dataset includes precise measures of EU struc-
tural funds by distinguishing between Objective 1, 2 and 3 payments over
the time period of 1999-2007.
To present an overview of the regional distribution of the structural funds,
Figure 1 shows the quantile maps displaying the distribution of the funds over
nine intervals by assigning the same number of values to each of the nine
categories in the map. The payments are expressed as a share of population
and are displayed as averages over the entire time period of observation. The
darker the area, the higher the share of that region’s payments of structural
funds per capita. The figures show that Ireland, Eastern Germany, Greece,
Southern Italy and Spain benefit most from Objective 1 payments, whereas
France, the UK, Northern Spain and Sweden show particularly high gains
from Objective 2 and Objective 3 payments, respectively.
We are not only interested in analysing the employment impact of to-
tal regional structural funds payments, we are also keen on distinguishing
between Objective 1, 2 and 3 payments. For this reason we start with spec-
ifications including the total sum of Objective 1+2+3 payments and then
Figure 1: Quantile maps, averages 1999–2006
Objective 1 payments Objective 2 payments
Objective 3 payments Objectives 1+2+3 payments
Notes: Own illustration. The payments of structural funds do not include multiregional funding pro-grammes. The darker the area, the higher the relative share of regions’ payments of structural funds percapita.
13ECB
Working Paper Series No 1403December 2011
14ECBWorking Paper Series No 1403December 2011
continue by investigating the impact of each single Objective. It could be
argued that structural funds projects, such as infrastructure investment, only
become effective after some time lag. Thus, we follow Mohl and Hagen (2010)
and analyse the impact of time lags in greater detail: We start our empirical
analysis by excluding any structural funds variable before gradually adding
the lagged structural funds variables; beginning with a lag of one year and
ending up with a specification comprising structural funds with lag of up to
four-years (sfi,t−j with j = 1, ..., 4).
Due to multicollinearity the coefficients and standard errors of the struc-
tural funds variable cannot be interpreted if the variable is included into
the regression with several lags. As a consequence, we calculate the sum
of structural funds coefficients (∑J
j=1 sfi,t−j) corresponding to the short-run
elasticity (Obj. short-term elast. (size)) and then use a simple Wald test to
determine whether the short-run elasticity is statistically different from zero
(Obj. short-term elast. (p-value)). As our estimation specification displayed
in equation (1) equals a dynamic approach, it is more convincing to inter-
pret the long-term impact of the structural funds. We do so and list its size
(Obj. long-term elast. (size)) and significance level (Obj. long-term elast.
(p-value)) in the regression output tables. The estimated long-term elasticity
could be used to show that a one per cent increase of structural funds (per
capita) leads to a rise of the regional employment level by X%.
Moreover, we provide a more parsimonious specification and control for
both country- and region-fixed effects by subtracting the annual country
mean from each of the variables instead of including dummy variables (Bond,
Hoeffler, and Temple, 2001). For variables (union density, gross replacement
rate) or countries (Denmark, Ireland and Luxembourg) where region-specific
variables are not available, we subtract the annual EU mean. For illustrative
purposes, the transformed employment level for Bavaria in year t is computed
by subtracting the German employment level in year t, whereas in the case of
Ireland, which only consists of one NUTS region, we subtract the EU mean
of year t.
15ECB
Working Paper Series No 1403December 2011
Furthermore, in order to avoid losing observations, we replace missing
entries of the compensation per employee variable with zero and include a
dummy variable which is equal to 1 if the variable contains a missing en-
try (for a similar approach see Fitzenberger, Kohn, and Wang, 2011). The
dummy variable is never statistically significant and thus not displayed in
the regression output tables. Finally, ui,t is the i.i.d. error term of the speci-
fication. Table 1 gives an overview of the precise definitions and data sources
of the variables used. The correlation matrix and the summary statistics are
displayed in Tables 2 and 3.
2.2 Spatial panel approach
The results of our baseline panel regression approach might be influenced
by regional spillover effects, which have been neglected so far, resulting in
biased estimates. In our sample of 130 European regions, the regions which
are located next to each other might disclose a stronger spatial dependence
than regions at a greater distance to one another.
In order to take these considerations into account, we apply spatial econo-
metric techniques, using a N ×N weight matrix (W ) containing information
about the connectivity between regions. Its diagonal consists of zeros, while
each wij specifies the way region i is spatially connected to region j. To
standardise the external influence upon each region, the weight matrix is
normalised so that the elements amount to one. We follow the approach
by Le Gallo and Ertur (2003) and Ertur and Koch (2006) and use a weight
matrix consisting of the k-nearest neighbours computed from the distance
between the centroids of the NUTS regions.3 This weight matrix is based
purely on geographical distance, which has the big advantage that exogeneity
3We use the Matlab toolbox “Arc Mat” (LeSage and Pace, 2004) to determine the centroids
of the polygons (regions) expressed in decimal degrees. These are converted to latitude
and longitude coordinates and are available upon request. The x nearest neighbours of
each region are then calculated with the help of the Spatial Statistics Toolbox 2.0 (Pace,
2003).
16ECBWorking Paper Series No 1403December 2011
of geographical distance is unambiguous. It is defined as follows:
W (k) =
w∗
ij(k) = 0 if i = j
w∗ij(k) = 1 if dij ≤ di(k) and wij(k) = w∗
ij(k)/∑
j w∗ij(k)
w∗ij(k) = 0 if dij > di(k)
where w∗ij is an element of the unstandardised weight matrix W and wij is
an element of the standardised weight matrix, di(k) is the smallest distance
of the kth order between regions i and j so that each region i has exactly k
neighbours.4
Generally, the inclusion of a spatially lagged dependent variable into a
panel fixed effects model generates an endogeneity problem because the spa-
tially weighted dependent variable is correlated with the disturbance term
(Elhorst, 2010). In order to control for this simultaneity, the following re-
sults are based on a quasi-maximum likelihood estimator for spatial dynamic
panel models as proposed by Yu, de Jong, and Lee (2008). This model
foresees spatially-weighted coefficients for both the lagged and the contem-
poraneous employment level. Apart from the inclusion of the spatial weight
variables, the selection of variables remains the same as in equation (1), so
we estimate the following model:
empi,t = β0 + λW empi,t + ρW empi,t−1 + γ empi,t−1 + β2 comp.empi,t−1+
+ β3 pop.youngi,t−1 + β4 low skilledi,t−1 + β5 market potentiali,t−1+
+ β6 grri,t−1 + β7 union densityi,t−1 + β8 sfi,t−j + µi + λt + ui,t
(2)
Unfortunately, it is currently not feasible to estimate a spatial lag model
and to control simultaneously for endogeneity of the other independent vari-
ables, for example with a (system) GMM approach. The reason for this
4For example, for k = 10 the elements of the row / column vector of the weight matrix
(W ) for the region “Region de Bruxelles-capitale” (be) are all zeros with the exception of
the ten nearest neighbours (be2, be3, fr10, fr21, fr22, fr30, fr41, nl2, nl3 and nl4) whose
elements are 0.1.
17ECB
Working Paper Series No 1403December 2011
is that introducing a spatial weight matrix creates a non-zero log-Jacobian
transformation from the disturbances of the model to the dependent variable,
while the system GMM procedure by Blundell and Bond (1998) is based on
the assumption of no Jacobian term involved.5
2.3 Panel approach with interaction term
As indicated in the introduction, it is not clear from a theoretical perspec-
tive whether EU funding is indeed associated with higher employment levels.
According to the capital-skill-complementary hypothesis (Griliches, 1969)
and the skill-based technological change hypothesis (Berman, Bound, and
Griliches, 1994) the demand for skilled labour increases with decreasing cap-
ital costs, while the demand for unskilled labour decreases with diminishing
capital costs. Hence, it might be argued that structural funds are only condi-
tionally effective depending on the regional education level. In order to test
this conditionality, we include an interaction term in the model of equation
(1) and estimate the following specification:
empi,t = β0 + β1 empi,t−1 + β2 comp.empi,t−1 + β3 pop.youngi,t−1+
+ β4 low skilledi,t−1 + β5 market potentiali,t−1 + β6 grri,t−1+
+ β7 union densityi,t−1 + β8 sfi,t−1 + β9 sfi,t−1 × low skilledi,t−1+
+ β10high skilledi,t−1 + µi + λt + ui,t
(3)
To interpret this model, we calculate the marginal effects of structural
funds on the employment level, which consists of the first derivative of the
above regression model (for a general overview on interaction models see
Braumoeller, 2004; Brambor, Clark, and Golder, 2006). This implies that
we have to evaluate the marginal effects at different values of low skilled. In
doing so, we take into account that the low-skilled variable is only defined
over a certain interval, and we calculate the marginal effects for a set of
5We thank James LeSage for his helpful advice.
18ECBWorking Paper Series No 1403December 2011
percentiles (5th, 10th..., 95th) between the minimum and the maximum of the
variable low skilled. In contrast to plotting the marginal effects over evenly
spaced values between the minimum and maximum of the low skilled vari-
able, the use of percentiles has the advantage that it illustrates the frequency
distribution of the variable and thus enables a more meaningful interpreta-
tion of the marginal effects. In addition, we indicate the level of uncertainty
regarding the marginal effects by plotting the lower and upper bound of the
95% confidence intervals. The details of the calculations are described in the
appendix in Section B.
3 Econometric results
From an econometric point of view, the investigation of employment effects
of EU funding poses several methodological challenges. First, the empiri-
cal results might be biased due to simultaneity: the allocation criteria of
the structural funds are likely to be correlated with the dependent variable
employment since its allocation depends, inter alia, on the regional unem-
ployment rate and the employment structure. Second, regional employment
variables might be influenced by regional spillover effects, as structural funds
payments may increase one region’s employment which, in turn, may af-
fect neighbouring regions’ employment positively or negatively. Finally, the
estimation results might strongly depend on the choice of the econometric
approach.
3.1 Baseline panel approach
We start with checking all specifications for autocorrelation using the test
proposed by Wooldridge (2002) (Table 5). As the Wooldridge test clearly
rejects the null hypothesis of no first-order autocorrelation, standard errors
are specified to be robust not only to heteroskedasticity but also to serial
19ECB
Working Paper Series No 1403December 2011
correlation as proposed by Newey and West (1987).6 We find a positive
and strongly significant impact of the lagged dependent variable. The size
and significance level of the coefficient hardly change, irrespective of how
many lags of the structural funds variable are included. As expected, our
wage variable (comp.emp) shows a negative coefficient, which is, however, not
significant. A high share of young population and of low level education leads
to a statistically significant reduction of the employment level. Moreover,
the regional market potential has a positive and significant impact on the
employment level. Both variables measuring labour market regulations at
the national level–the gross replacement rate and the union density–are not
statistically significant.
The main variable of interest is the structural funds variable. Table 5
reveals that the total Objective 1+2+3 payments are not statistically signif-
icant. One reason for this might be that the estimation results are biased
due to endogeneity of the structural funds variable, since the employment
structure is one criterion for the allocation of structural funds. In order
to deal with this issue, the literature has suggested two kinds of external
instrument variables. Dall’erba and Le Gallo (2008) instrument structural
funds payments by the regions’ distance to Brussels, arguing that the spatial
distribution of structural funds payments follows a centre-periphery pattern.
Bouvet (2005) uses partisan affinity as an instrument for structural funds.
However, while the first set of instruments shows no variation over time at all,
the time variation of variables related to political affinity is low and in some
regions even zero. Thus, their effect on structural funds payments is absorbed
once regional fixed effects are controlled for, rendering them unsuitable for a
panel fixed effects approach.
As a consequence, we address the problem of endogeneity by basing the
identification on internal instruments via a system GMM estimator (Blun-
6As a robustness check, we used the estimation procedure proposed by Prais-Winsten
and Driscoll and Kraay (1998). The results hardly change and they are available upon
request.
20ECBWorking Paper Series No 1403December 2011
dell and Bond, 1998). We assume that lagged employment, compensation
per employee, education, market potential and structural funds payments
are endogenous. The standard errors are finite-sample adjusted following
Windmeijer (2005). When using the system GMM estimator the number of
instruments grows quadratically with T . Too many instruments can overfit
the instrumented variables (Roodman, 2009), reduce the power properties of
the Hansen test (Bowsher, 2002) and lead to a downward-bias in two-step
standard errors (Windmeijer, 2005). In order to guarantee a parsimonious
use of instruments, we follow Mehrhoff (2009) and limit the number of instru-
ments by using the ‘collapse’ option Roodman (2009). As a robustness check
we also increased the number of instruments in the system GMM regressions;
however, the results hardly differ.
Given this parsimonious specification, the estimation results show that
the Hansen test of overidentifying restrictions is not statistically significant,
i.e. the null hypothesis which states that the instruments are not correlated
with the residual cannot be rejected (Table 5). We also report the p-values
for the tests of serial correlation. These tests are based on first-differenced
residuals and we expect the disturbances ui,t not to be serially-correlated in
order to yield valid estimation results. The regression output in Table 5 shows
no second-order serial correlation (AR(2) (p-value)). For most variables,
the size and significance level are comparable to the results of the previous
regressions. The use of the system GMM estimator slightly increases the size
of the coefficients of the lagged dependent variable, while the market potential
variable is no longer statistically significant. Above all, the Objective 1+2+3
variable is still not statistically significant.
Even though the total payments of structural funds have no significant
impact, it cannot be ruled out that sub-parts of the EU funding significantly
affect the employment level. As a consequence, we re-run our regression
model using more precise measures of structural funds, distinguishing be-
tween Objective 1, 2 and 3 payments. The results show that the size of the
coefficients of the Newey and West specifications are in line with the results
21ECB
Working Paper Series No 1403December 2011
of the more aggregated analysis (Table 6). In particular, the coefficients of
the disaggregated structural funds variable are not statistically significant.
Switching to the system GMM estimator again slightly increases the size of
the coefficients of the lagged dependent variable (Table 6). Moreover, the
short- and long-term elasticities of Objective 1 payments now show jointly
statistically significant negative coefficients when the structural funds vari-
able is included with more than one lag, while Objective 3 payments have a
significantly positive coefficient when more than two lags are included.
As mentioned above, the most likely channel through which structural
funds affect employment is an increase in the regional capital endowment,
which leads to an increase of the marginal product of labour, the output
level, and, ceteris paribus, the labour demand, given a matching labour sup-
ply. When estimating the effect of EU funding on employment, some part
of the causal effect might be, at least in an indirect way, absorbed by the
inclusion of the market potential. For this reason we replace our proxy for
the output level and define market potential 2 for region i as the weighted
GDP of the neighbouring regions, thereby excluding the GDP of region i.
The reduced-form approach including the regions’ output level may be inter-
preted as being based on a ‘conditional labour demand model’, the estimation
strategy without the regions’ output level as being based on an ‘unconditional
labour demand model’.
In line with the results of described above, the size and significance level
of the independent variables hardly change.7 In particular, our indicator
measuring market potential is still positive and the total structural funds
variable is not significant. We also estimated the model using the disag-
gregated structural funds variables. The size and significance levels remain
broadly unchanged except that the Objective 3 variable is no longer statisti-
cally significant.
The use of the market potential 2 variable is still associated with the
potential problem of regional spillover effects. As a consequence, we drop
7The detailed regression results are available upon request.
22ECBWorking Paper Series No 1403December 2011
the market potential variable and re-run the regressions. Table 7 reveals that
the results of the independent variables broadly remain unchanged and that
the structural funds variable is still not statistically significant. Switching to
the disaggregated analyses shows that Objective 1 payments partly show a
negative and significant coefficient, while Objective 3 funding has a jointly
significant positive impact if the structural funds variable is included with
more than one lag.
3.2 Spatial panel approach
The estimation of a spatial panel model requires the definition of a spatial
weight matrix. We start our regression analysis with a very low value for
the indicator measuring the closeness; assuming that the spillover effects are
limited to the four closest regions (k = 4). Table 8 reveals that the contem-
poraneous spatial weight matrix (γ) has a positive and strongly significant
coefficient, while the lagged spatial weight matrix (ρ) has a negative and
statistically significant coefficient. This implies that spillover effects seem to
have an immediate positive cross-regional effect, boosting the employment
level before they turn negative. This negative impact may be explained by
migration and commuting, i.e. people tend to move or commute to the neigh-
bouring regions if economic differences of the regional labour market persist,
resulting in negative employment effects in the origin region.
Apart from the spillover effect, the results show that the significance
levels of the coefficients are broadly comparable with the non-spatial regres-
sions. The size of most coefficients is slightly reduced as some of the causal
relationship can be explained by regional spillover effects. The lagged de-
pendent variable still has a strong positive impact on the employment level.
A high share of young population and low levels of education have a signifi-
cantly negative effect. Market potential promotes the regional employment,
and the coefficients of union density and the gross replacement rate are not
statistically significant.
As regards the structural funds variable, Table 8 reveals that total struc-
23ECB
Working Paper Series No 1403December 2011
tural funds now seem to have a jointly negative impact if more than two
lags are included. Using more disaggregated structural funds data, we find a
small negative impact of Objective 3 payments (Table 9). The results do not
change when switching to the model excluding the market potential variable
(Tables 8, 9).
As some papers claim that the regression results are very sensitive to the
choice of the weight matrix (LeSage and Fischer, 2008; LeSage and Pace,
2010), we also estimate our regression model for various spatial weight ma-
trices, i.e. we use different parameters of k, an inverse euklidean (W.dist)
and an inverse squared euklidean (W.dist2) distance weight matrix. Table
10 shows that with a larger coefficient of the weight matrix the coefficients
of the contemporaneous and lagged weight matrices rise. However, the in-
creases are limited to a certain range, and the size and significance levels of
the other independent variables are not substantially affected. Irrespective
of the choice of the matrix, the weight coefficients are statistically significant
at the 1% level. Furthermore, the size and significance levels of the other
independent variables hardly change with a different weight matrix.
3.3 Panel approach with interaction term
Finally, we investigate whether structural funds are conditionally effective
depending on the education levels of the working age population, i.e. the
skill-level of labour supply. For this purpose, we estimate an interaction
model using the structural funds and the low-skill variable in an interac-
tion framework. Unlike the remaining independent variables, the low-skilled
variable is only available from the year 1999 onwards, so we restrict our
estimation to three lags only.
The results displayed in Table 11 show that the lagged dependent variable
is still strongly significant, while the statistical significance of the remaining
coefficients is reduced. The coefficient of the interaction term tells us how
the marginal effect varies according to values of low education, while its
significance level tests whether low education (linearly) conditions the effect
24ECBWorking Paper Series No 1403December 2011
of structural funds on employment (and vice versa). However, as indicated
above, it is more convincing to base the interpretation on the calculation of
the marginal effects.
We graph the marginal effects of the short- and long-term elasticities for
varying values of the education variable, starting with the total structural
funds (top left panel) and followed by Objective 1 (top right panel), Objective
2 (bottom left panel) and Objective 3 (bottom right panel) payments (Figure
2). The straight line displayed in the graphs represent the marginal effect
of structural funds surrounded by 95% confidence intervals. The marginal
effects of structural funds are a linear function of low skilled. Moreover,
the coefficients displayed in Table 11 indicate the impact of structural funds
when low skilled is zero, while the interaction coefficient gives our estimate
of the slope of the marginal effect line.
Figure 2 shows that the marginal effects of structural funds and our con-
fidence regarding the marginal effects vary with values of low skilled. More-
over, the marginal effects of total structural funds payments clearly show a
negative slope. The total structural funds payments have a positive impact
on the employment level in regions with a low share of low-skilled population,
while they have a negative impact in regions with a high share of low-skilled
population. These insights are particularly valid for the marginal effects of
the long-term elasticities of Objective 1+2+3 and of Objective 1 payments.
As regards Objective 2 payments, the slope of the marginal effects depends
on the number of lags and the confidence intervals point to no significant
impact. Finally, the marginal effects of the Objective 3 payments have a
slight negative slope but do not turn negative.
These results are still valid when switching to the model excluding mar-
ket potential and when including the high-skilled variable as an additional
independent variable. Moreover, the results hardly change when estimating a
dynamic spatial panel interaction model in the model including or excluding
market potential.8 Finally, we estimate our interaction model by replacing
8The estimation results are not displayed in their entirety due to space constraints but
25ECB
Working Paper Series No 1403December 2011
the variable measuring educational attainment. We interact the structural
funds variable with an indicator measuring the share of high-skilled popula-
tion. Figure 3 illustrates that this leads to a positive linear effect, implying
that structural funds have a positive impact on the employment level in re-
gions with a high share of high-skilled population, whereas they negatively
affect the employment level in regions with a low share of high-skilled popu-
lation.
4 Conclusions
While the current literature on the effectiveness of EU funding has primarily
concentrated on the investigation of the economic growth effects, the aim
of this paper is to evaluate their employment impact. From a theoretical
perspective higher expenditures on EU funding do not necessarily lead to
higher total employment levels. Instead, its effectiveness depends, in partic-
ular, on whether structural funds payments are used as capital subsidies or
as human capital investment and it is subject to the educational attainment
of the population as well as to the labour market tightness. The paper con-
tributes to the literature by (i) investigating the relevance of the inconclusive
theoretical prediction via the estimation of interaction effects, (ii) analysing
more precise measures of EU aid over an extended time period and (iii) ap-
plying dynamic (spatial) panel techniques, controlling for heteroscedasticity,
serial and spatial correlation, as well as for endogeneity. In particular, using
a spatial dynamic panel approach, we find that regional spillovers do have
a significant impact on the regional employment level irrespective of which
Objective and time lag is analysed.
In line with the theoretical predictions, we find no clear evidence that
are available upon request. We also estimated the regression model with various spatial
weight matrices in order to check the robustness of the results. The empirical evidence,
which is available upon request, shows that the spatial panel interaction model does not
depend on the choice of the spatial weight matrix.
26ECBWorking Paper Series No 1403December 2011
EU funding promotes employment. Instead, structural funds payments seem
to be used as capital subsidies: they have a statistically positive impact on
employment in regions with a low share of low-skilled population, and they
have a negative impact on the employment level in regions with a high share
of low-skilled population. Broadly summarising, we find that a one per cent
increase of total structural funds payments leads to a positive (negative)
impact on the regional employment by approximately 0.05% in regions with
a high (low) share of skilled population. These results seem to be mainly
driven by Objective 1 funding, which corresponds to the largest part of total
structural funds payments.
Apart from the theoretically-founded explanation, a statistically insignifi-
cant, or even negative, impact of structural funds payments can be explained
by at least four factors: First, in contrast to Objective 1 payments, Objec-
tive 2 and 3 payments are not solely based on clear criteria. Hence, there is
room for political bargaining and/or side payments so that politically mo-
tivated projects are financed rather than economically efficient and growth-
increasing projects. Second, de jure the structural funds payments have to
be co-financed. However, recent panel studies using country data provide ev-
idence that some crowding out of national public investment may take place
(Hagen and Mohl, 2011a). This, in turn, might have a negative impact on
the regional GDP. Third, Cohesion policy could be ineffective with regard
to human capital investment. Finally, a positive employment effect due to
additional labour demand driven by a short-term aggregate demand stimulus
is only possible if the quality and quantity of labour supply suffices. This
may not be the case in periods of positive output gaps, for example, in the
new East-European member states.
The results have policy implications for the setup of future Multiannual
Financial Frameworks. It becomes evident that EU funding lacks a clear
concept on how to promote employment in the medium- to long-run. Our
results indicate that the high-skilled population in particular benefits from
EU structural funds payments. As a consequence, a strategy should define
27ECB
Working Paper Series No 1403December 2011
objectives which are clearly measurable and allow for an ex-post assessment
of this policy field. This, in turn, would contribute to a more effective policy.
Acknowledgements
We thank Theodor Martens, Gabriel Glockler, Stefan Huemer, Heidi Hellerich, one anony-
mous referee and the Editorial Board of the ECB Working Paper Series for providing
fruitful comments.
28ECBWorking Paper Series No 1403December 2011
Appendix
A Description of the dataset
The European regions are classified by the European Commission into threedifferent groups called “Nomenclature des unites territoriales statistiques”(NUTS). These units refer to the country level (NUTS-0) and to three lowersubdivisions (NUTS-1, NUTS-2 and NUTS-3), which are classified accordingto the size of population. Our dataset consists of both NUTS-1 and NUTS-2 regions. In order to guarantee the highest degree of transparency, thissection lists the abbreviations of the NUTS codes in brackets following theclassifications of the European Commission (2007).
The choice of the NUTS level follows the data availability of structuralfunds payments. Generally, we try to use data on NUTS-2 level wheneverpossible. This is the case for France, Greece, Italy, Portugal, Spain, andSweden. However, in case of Germany we have to use NUTS-1 level becausethe annual reports do not contain more detailed information. Moreover, insome countries there is no clear-cut distinction in the sense that in the annualreports the structural funds are partly allocated to the NUTS-1 and partly tothe NUTS-2 level. Finally, the annual reports of structural funds for 1995 and1996 (European Commission, 1996b, 1997) for some countries only containdata at the NUTS-1 level. As a consequence, we chose the NUTS-1 level forAustria, Belgium, Finland, the Netherlands, and the United Kingdom.
For Denmark and Luxembourg subdivisions do not exist, so that NUTS-0,NUTS-1 and NUTS-2 codes are the same. We regard these cases as NUTS-2regions. In Ireland the labels of NUTS-0 and NUTS-1 are identical, so thatwe classify Ireland as a NUTS-1 region. Please note that we do not considerthe overseas regions of France (Departments d’outre-mer (fr9) consisting ofGuadeloupe (fr91), Martinique (fr92), Guyane (fr93) and Reunion (fr94)),Portugal (Regiao Autonoma dos Acores (pt2, pt20), Regiao Autonoma daMadeira (pt3, pt30)), and Spain (Canarias (es7, es70)). As a consequence,our dataset consists of the following 130 NUTS-1 and NUTS-2 regions, forwhich we have structural funds payments:
Belgium (3 NUTS-1 regions): Region de Bruxelles-capitale (be1), VlaamsGewest (be2), Region Wallonne (be3);
Denmark (1 NUTS-2 region): Denmark (dk);Germany (16 NUTS-1 regions): Baden-Wurttemberg (de1), Bayern (de2),
Berlin (de3), Brandenburg (de4), Bremen (de5), Hamburg (de6), Hessen (de7),Mecklenburg-Vorpommern (de8), Niedersachsen (de9), Nordrhein-Westfalen (dea),
29ECB
Working Paper Series No 1403December 2011
Rheinland-Pfalz (deb), Saarland (dec), Sachsen (ded), Sachsen-Anhalt (dee), Schles-wig-Holstein (def), Thuringen (deg);
Greece (13 NUTS-2 regions): Anatoliki Makedonia, Thraki (gr11), Ken-triki Makedonia (gr12), Dytiki Makedonia (gr13), Thessalia (gr14), Ipeiros (gr21),Dytiki Ellada (gr23), Ionia Nisia (gr22), Sterea Ellada (gr24), Peloponnisos (gr25),Attiki (gr30), Voreio Aigaio (gr41), Notio Aigaio (gr42), Kriti (gr43);
Spain (16 NUTS-2 regions): Galicia (es11), Principado de Asturias (es12),Cantabria (es13), Paıs Vasco (es21), Comunidad Foral de Navarra (es22), La Ri-oja (es23), Aragon (es24), Comunidad de Madrid (es30), Castilla y Leon (es41),Castilla-La Mancha (es42), Extremadura (es43), Cataluna (es51), Comunidad deValenciana (es52), Illes Balears (es53), Andalucıa (es61), Region de Murcia (es62),Ciudad Autonoma de Ceuta (es63), Ciudad Autonoma de Melilla (es64);
France (22 NUTS-2 regions): Ile de France (fr10), Champagne-Ardenne (fr21),Picardie (fr22), Haute-Normandie (fr23), Centre (fr24), Basse-Normandie (fr25),Bourgogne (fr26), Nord-Pas-de-Calais (fr30), Lorraine (fr41), Alsace (fr42),Franche-Comte (fr43), Pays-de-la-Loire (fr51), Bretagne (fr52), Poitou-Charentes(fr53), Aquitaine (fr61), Midi-Pyrenees (fr62), Limousin (fr63), Rhone-Alpes (fr71),Auvergne (fr72), Languedoc-Roussillon (fr81), Provence-Alpes-Cote d’Azur (fr82),Corse (fr83);
Ireland (1 NUTS-1 region): Irland (ie);Italy (21 NUTS-2 regions): Piemonte (itc1), Valle d’Aosta/Vallee d’Aoste
(itc2), Liguria (itc3), Lombardia (itc4), Provincia autonoma Bolzano (itd1), Provin-cia autonoma Trento (itd2), Veneto (itd3), Friuli-Venezia Giulia (itd4), Emilia-Romagna (itd5), Toscana (ite1), Umbria (ite2), Marche (ite3), Lazio (ite4),Abruzzo (itf1), Molise (itf2), Campania (itf3), Puglia (itf4), Basilicata (itf5), Cal-abria (itf6), Sicilia (itg1), Sardegna (itg2);
The Netherlands (4 NUTS-1 regions): Noord-Nederland (nl1), Oost-Neder-land (nl2), West-Nederland (nl3), Zuid-Nederland (nl4);
Luxembourg (1 NUTS-1 region): Luxembourg (lu);Austria (3 NUTS-1 regions): Ostosterreich (at1), Sudosterreich (at2), West-
osterreich (at3);Portugal (5 NUTS-2 regions): Norte (pt11), Algarve (pt15), Centro (P)
(pt16), Lisboa (pt17), Alentejo (pt18);Finland (2 NUTS-1 regions): Manner-Suomi (fi1), Aland (fi2);Sweden (8 NUTS-2 regions): Stockholm (se11), Ostra Mellansverige (se12),
Smaland med oarna (se021), Sydsverige (se22), Vastsverige (se23), Norra Mel-lansverige (se31), Mellersta Norrland (se32), Ovre Norrland (se33);
UK (12 NUTS-1 regions): North East (ukc), North West (ukd), Yorkshire and
the Humber (uke), East Midlands (ukf), West Midlands (ukg), East of England
(ukh), London (uki), South East (ukj), South West (ukk), Wales (ukl), Scotland
(ukm), Northern Ireland (ukn).
30ECBWorking Paper Series No 1403December 2011
B Descriptive statistics and regression results
Table 1: Variables and data sources
Variable Definition Source
Emp Employment (reg lfe2enace) over totalpopulation between 15 and 64 years (reg d2avg)
Comp. emp Compensation of employees in the manufacturingsector in million of Euro (reg e2rem)
Pop. young Share of population aged 15 and below (reg d2jan)over total population
Low-skilled Share of population aged 15 and over whose highestlevel of education is pre-primary, primary and lowersecondary education – levels 0-2 according to the Eurostat Regio statistics (the
official Eurostat codes are listed inparentheses)
International Standard Classification of Education(ISCED) 1997 (reg lfsd2pedu)
High-skilled Share of population aged 15 and over whose highestlevel of education is tertiary education – levels 5-6according to ISCED (1997) (reg lfsd2pedu)
Market potential Sum of GDP (reg e2gdp) of region i and GDP of allother regions k, weighted by the square of theEuclidean distance from region i to region k
market potentiali,t = GDPi,t +∑
k(GDPk/d2ik).
Market potential 2 GDP of all other regions k, weighted by the squareof the Euclidean distance from region i to region k
market potential2i,t =∑
k(GDPk/d2ik).
grr Gross replacement rate, which measures the average ofthe gross unemployment benefit replacement rates for OECD, database on unemployment
benefit entitlements andreplacement rates
two earnings levels, three family situations and threedurations of unemployment divided by 100. The originaldata are for every second year and have been linearlyinterpolated.
Union density Trade union density OECD
SF pc Obj. 1 Objective 1 payments per capita in Euro Data for 1999 are from theEuropean Commission (2000);Data for the period 2000–2006were accessed at the EuropeanCommission in Brussels on 24/25November 2007
SF pc Obj. 2 Objective 2 payments per capita in Euro
SF pc Obj. 3 Objective 3 payments per capita in Euro
SF pc Obj. 1+2+3 Objectives 1+2+3 payments per capita in Euro
Table 2: Correlation matrix
Emp. Comp. Pop. Low- Grr Unionemp. young skilled density
Emp. 1Comp. emp. 0.4411 1Pop. young 0.1267 0.3352 1Low-skilled -0.3511 -0.7035 -0.2066 1Grr 0.3549 0.6781 0.2533 -0.4792 1Union density 0.3767 0.6633 0.3282 -0.5468 0.6657 1Market potential 0.0298 0.0821 0.097 0.2428 -0.0713 -0.0632Market potential 2 0.3388 0.1172 0.0789 -0.3231 -0.0269 -0.0304SF pc Obj. 1 -0.403 -0.452 -0.2357 0.2999 -0.3702 -0.3348SF pc Obj. 2 0.0757 0.1218 0.0103 -0.088 -0.0196 0.1219SF pc Obj. 3 0.0663 -0.1118 -0.2965 0.1362 -0.0372 -0.0018SF pc Obj. 1+2+3 -0.4062 -0.4644 -0.2778 0.3121 -0.406 -0.3284
Market Market SF pc SF pc SF pc SF pcpotential potential 2 Obj. 1 Obj. 2 Obj. 3 Obj. 1+2+3
Market potential 1Market potential 2 -0.0431 1SF pc Obj. 1 -0.144 -0.0851 1SF pc Obj. 2 0.1441 -0.0935 -0.3468 1SF pc Obj. 3 0.0923 0.1068 -0.2243 0.2284 1SF pc Obj. 1+2+3 -0.1092 -0.1047 0.9637 -0.0989 -0.088 1
31ECB
Working Paper Series No 1403December 2011
Table 3: Summary statistics
Variable Mean Std. dev. Minimum Maximum Observations
Emp. overall 0.6334 0.0833 0.2456 0.8398 N = 1142between 0.0803 0.3669 0.8253 n = 130within 0.0244 0.5121 0.7119 T = 8.7846
Comp. emp. overall 24,377.8 8,389.6 5,899.5 47,353.4 N = 961between 8,585.8 7,504.2 47,080.6 n = 130within 1,490.1 15,193.8 30,414.2 T = 7.3923
Pop. young overall 0.1608 0.0261 0.1000 0.2340 N = 1163between 0.0257 0.1034 0.2249 n = 130within 0.0054 0.1408 0.1888 T = 8.9462
Low-skilled overall 0.4443 0.1789 0.0911 0.8746 N = 1163between 0.1669 0.1557 0.8352 n = 130within 0.0671 0.0133 0.6447 T = 8.9462
Grr overall 0.3127 0.1012 0.1207 0.6107 N = 1168between 0.0986 0.1329 0.5166 n = 130within 0.0238 0.1954 0.4068 T = 8.9846
Union density overall 0.2779 0.1813 0.0782 0.8063 N = 1106between 0.1775 0.0813 0.7705 n = 130within 0.0124 0.2151 0.3193 T = 8.5077
Market potential overall 155,565.6 109,534.1 22,697.7 606,570.4 N = 1032between 109,565.2 24,062.4 595,974.7 n = 129within 8,641.8 78,995.9 207,308.0 T = 8
Market potential 2 overall 78,719.1 37,566.6 16,052.9 282,920.6 N = 1032between 37,495.7 16,687.0 257,092.1 n = 129within 3,855.2 43,116.4 104,547.6 T = 8
SF pc Obj. 1 overall 44.2765 71.8250 0.0000 408.1175 N = 1169between 63.0015 0.0000 272.1494 n = 130within 34.9174 0.0000 280.3748 T = 8.9923
SF pc Obj. 2 overall 10.2749 16.8297 0.0000 300.6687 N = 1170between 11.9288 0.0000 67.8109 n = 130within 11.9129 0.0000 243.1328 T = 9
SF pc Obj. 3 overall 2.5094 6.3707 0.0000 51.3384 N = 1170between 5.6338 0.0000 33.4844 n = 130within 3.0106 0.0000 23.9043 T = 9
SF pc Obj. 1+2+3 overall 57.0718 66.9904 0.0000 408.1175 N = 1169between 56.0263 0.0000 272.1494 n = 130within 37.0368 0.0000 298.6794 T = 8.9923
Table 4: Definition of the structural funds variables by Objective, 1994–2006
1994-1999 2000-2006
Definition share of Definition share oftotal SF total SF
Obj. 1: To promote the development and struc-67.6%
Obj. 1: Supporting development in the69.7%
tural adjustment of regions whose development less prosperous regionsis lagging behind the rest of the EUObj. 6: Assisting the development of sparsely-
0.5%populated regions (Sweden & Finland only)
Obj. 2: To convert regions seriously affected11.1%
Obj. 2: To support the economic and11.5%by industrial decline social conversion of areas experiencing
Obj. 5b: Facilitating the development and4.9%
structural difficultiesstructural adjustment of rural areas
Obj. 3: To combat long-term unemployment &
10.9%
Obj. 3: To support the adaptation and mo-
12.3%facilitate the integration into working life of dernisation of education, training & employ-young people & of persons exposed to ex- ment policies in regions not eligible underclusion from the labour market Obj. 1Obj. 4: To facilitate the adaptation of workersto industrial changes and to changes in produc-tion systems
Source: European Commission.
32ECBWorking Paper Series No 1403December 2011
Table
5:Reduced-form
employmentmodel
includingmarketpotential:Objectives
1+2+
3
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Newey
and
West
(1987)
Two-ste
psy
stem
GM
M
Emp.perwp.(t-1
)0.469***
0.469***
0.468***
0.466***
0.467***
0.593***
0.594***
0.664***
0.718***
0.737***
(4.063)
(4.060)
(4.095)
(4.077)
(4.083)
(6.738)
(7.799)
(9.635)
(9.830)
(11.21)
Comp.emp.(t-1
)-0
.00329
-0.00331
-0.00237
-0.00252
-0.00252
-0.00392
-0.00529
-0.0349
-0.0303
-0.0301
(-0.241)
(-0.242)
(-0.175)
(-0.185)
(-0.185)
(-0.0596)
(-0.112)
(-0.779)
(-0.852)
(-0.711)
Pop.young
(t-1
)-0
.156***
-0.155***
-0.157***
-0.158***
-0.158***
-0.113**
-0.120**
-0.0858*
-0.0817*
-0.0812*
(-4.381)
(-4.361)
(-4.399)
(-4.384)
(-4.383)
(-2.586)
(-3.072)
(-2.174)
(-2.157)
(-2.163)
Low-skilled
(t-1
)-0
.0749*
-0.0750*
-0.0725*
-0.0724*
-0.0726*
-0.0632
-0.0696*
-0.100**
-0.0701
-0.0645
(-2.010)
(-2.012)
(-1.990)
(-1.995)
(-1.991)
(-1.762)
(-2.326)
(-2.658)
(-1.694)
(-1.745)
Grr
(t-1
)0.000342
0.000346
0.000199
0.000155
0.000172
-0.000719
0.00163
0.00247
0.00173
0.000825
(0.0362)
(0.0366)
(0.0203)
(0.0157)
(0.0175)
(-0.125)
(0.317)
(0.416)
(0.348)
(0.197)
Union
density
(t-1
)0.00410
0.00405
0.00542
0.00609
0.00599
-0.00230
-0.00231
-0.00119
-0.00195
-0.00195
(0.128)
(0.126)
(0.168)
(0.187)
(0.183)
(-0.580)
(-0.571)
(-0.302)
(-0.466)
(-0.627)
Mark
etpote
ntial(t-1
)0.122*
0.123*
0.105*
0.100
0.0999
0.0542
0.0483
0.0416
0.0309
0.0360
(2.449)
(2.472)
(2.063)
(1.946)
(1.945)
(1.062)
(1.337)
(0.951)
(1.164)
(1.377)
SF
pcObj.
1+2+3
(t-1
)0.000177
0.000437
0.000453
0.000451
-0.00236
-0.00258
-0.00486
-0.00519
(0.0968)
(0.240)
(0.249)
(0.248)
(-0.436)
(-0.523)
(-0.991)
(-1.074)
SF
pcObj.
1+2+3
(t-2
)-0
.00347*
-0.00336*
-0.00337*
7.94e-0
50.000337
3.91e-0
5(-2.045)
(-2.040)
(-2.057)
(0.0396)
(0.124)
(0.0165)
SF
pcObj.
1+2+3
(t-3
)-0
.00147
-0.00149
0.000787
0.000542
(-0.941)
(-0.945)
(0.387)
(0.316)
SF
pcObj.
1+2+3
(t-4
)0.000237
0.00213
(0.214)
(1.269)
Constant
-0.117**
-0.117**
-0.100
-0.0970
-0.0960
0.00213
0.00226
0.00233
0.00169
0.00139
(-2.599)
(-2.598)
(-1.436)
(-1.358)
(-1.350)
(0.790)
(0.842)
(0.869)
(0.830)
(0.796)
Obj.
1+2+3
short-term
elast.(size)
-0.00303
-0.00437
-0.00417
-0.00250
-0.00374
-0.00248
Obj.
1+2+3
short-term
elast.(p
-valu
e)
0.153
0.103
0.144
0.599
0.545
0.634
Obj.
1+2+3
long-term
elast.(size)
0.000334
-0.00571
-0.00819
-0.00783
-0.00582
-0.00743
-0.0133
-0.00942
Obj.
1+2+3
long-term
elast.(p
-valu
e)
0.923
0.152
0.105
0.154
0.666
0.607
0.519
0.618
Woold
ridgeAR(1
)(p
-valu
e)
00
00
0AR(1
)(p
-valu
e)
0.00243
0.00168
0.000473
0.000342
0.000273
AR(2
)(p
-valu
e)
0.248
0.246
0.265
0.278
0.300
Hanse
n(p
-valu
e)
0.422
0.455
0.449
0.392
0.645
No.ofin
stru
ments
33
41
49
57
65
No.ofobse
rvations
959
959
959
959
959
959
959
959
959
959
No.ofre
gions
129
129
129
129
129
129
129
129
129
129
Notes:In
colu
mns(1
)to
(5)standard
errors
are
calculate
daccord
ingto
Newey
and
West
(1987),
t-statisticsare
reported
inpare
nth
ese
s.In
colu
mns(6
)to
(10)z-sta
tistics
are
listed
inpare
nth
ese
sapplyin
gth
etw
o-ste
psy
stem
GM
Mestim
ato
ras
pro
pose
dby
(Blu
ndell
and
Bond,1998).
The
lagged
dependent
variable,com
pensa
tion
per
employee,low-skilled,mark
etpote
ntialand
thestru
ctu
ralfu
ndsvariablesare
assumed
tobeendogenous.
Wein
stru
mentth
eendogenousvariableswith
both
itslagsand
itsdiff
ere
nced
lagsand
use
the“collapse
”option.Sta
ndard
errors
are
correcte
dusing
theappro
ach
by
Win
dm
eijer(2
005).
*signifi
cantat10%
;**
signifi
cantat5%
;***
signifi
cantat1%.
33ECB
Working Paper Series No 1403December 2011
Table
6:Reduced-form
employmentmodel
includingmarketpotential:Objectives
1,2,
3
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
Newey
and
West
(1987)
Two-ste
psy
stem
GM
M
Emp.perwp.(t-1
)0.469***
0.467***
0.468***
0.467***
0.468***
0.588***
0.629***
0.670***
0.677***
(4.063)
(4.026)
(4.045)
(4.034)
(4.040)
(8.791)
(11.39)
(14.18)
(16.92)
Comp.emp.(t-1
)-0
.00329
-0.00368
-0.00336
-0.00363
-0.00352
-0.0503
-0.0760*
-0.0587*
-0.0509
(-0.241)
(-0.268)
(-0.244)
(-0.263)
(-0.255)
(-1.049)
(-1.973)
(-2.208)
(-1.190)
Pop.young
(t-1
)-0
.156***
-0.154***
-0.151***
-0.151***
-0.150***
-0.104*
-0.0658
-0.0471*
-0.0258
(-4.381)
(-4.296)
(-4.154)
(-4.053)
(-4.003)
(-2.532)
(-1.657)
(-2.069)
(-0.960)
Low-skilled
(t-1
)-0
.0749*
-0.0746*
-0.0722
-0.0707
-0.0720
-0.0947**
-0.103***
-0.0746
-0.0635
(-2.010)
(-1.988)
(-1.946)
(-1.900)
(-1.885)
(-2.688)
(-3.439)
(-1.754)
(-1.657)
Grr
(t-1
)0.000342
0.000384
0.000125
0.000167
0.000228
-0.00180
-0.000712
-0.00277
-0.00155
(0.0362)
(0.0409)
(0.0130)
(0.0172)
(0.0236)
(-0.265)
(-0.114)
(-0.547)
(-0.322)
Union
density
(t-1
)0.00410
0.00417
0.00542
0.00593
0.00570
-0.00237
-0.00239
-0.00218
-0.00161
(0.128)
(0.130)
(0.166)
(0.180)
(0.173)
(-0.547)
(-0.608)
(-0.534)
(-0.335)
Mark
etpote
ntial(t-1
)0.122*
0.125*
0.119*
0.116*
0.115*
0.0599
0.0359
0.0215
0.0196
(2.449)
(2.502)
(2.346)
(2.255)
(2.288)
(1.272)
(1.191)
(1.055)
(1.145)
SF
pcObj.
1(t-1
)0.000650
0.00156
0.00150
0.00143
-0.00394
-0.00416
-0.00692
-0.00632
(0.452)
(1.082)
(1.058)
(1.018)
(-0.866)
(-1.266)
(-1.539)
(-1.784)
SF
pcObj.
1(t-2
)-0
.00363*
-0.00389*
-0.00398*
-0.00234
-0.00186
-0.00216
(-2.140)
(-2.336)
(-2.311)
(-1.215)
(-0.682)
(-0.888)
SF
pcObj.
1(t-3
)-0
.000244
-0.000316
0.000920
0.000875
(-0.177)
(-0.222)
(0.372)
(0.452)
SF
pcObj.
1(t-4
)0.000579
0.000953
(0.389)
(0.635)
SF
pcObj.
2(t-1
)-0
.00116
-0.00198
-0.00192
-0.00185
0.000992
0.00399
0.00813
0.00527
(-0.603)
(-1.049)
(-1.000)
(-0.948)
(0.204)
(0.535)
(1.043)
(0.804)
SF
pcObj.
2(t-2
)0.00105
0.00120
0.00114
0.000819
-0.000518
0.000460
(0.726)
(0.833)
(0.766)
(0.266)
(-0.126)
(0.149)
SF
pcObj.
2(t-3
)-0
.00181
-0.00174
-0.00134
-0.000178
(-1.195)
(-1.170)
(-0.542)
(-0.0922)
SF
pcObj.
2(t-4
)0.000365
0.00164
(0.264)
(1.132)
SF
pcObj.
3(t-1
)0.00131
0.00191
0.00178
0.00209
0.00546
0.00994
0.00998
0.0180**
(0.662)
(0.946)
(0.891)
(1.026)
(0.692)
(0.906)
(1.570)
(2.641)
SF
pcObj.
3(t-2
)-0
.000654
-0.000836
-0.000665
-0.00367
-0.000984
-0.00177
(-0.378)
(-0.498)
(-0.400)
(-0.971)
(-0.196)
(-0.304)
SF
pcObj.
3(t-3
)0.000586
0.000751
0.00455
0.00384
(0.368)
(0.469)
(1.547)
(1.192)
SF
pcObj.
3(t-4
)-0
.00119
-0.00374
(-0.779)
(-1.027)
Constant
-0.117**
-0.122**
-0.114**
-0.113
-0.112
0.00259
0.00254
0.00171
0.000849
(-2.599)
(-3.089)
(-2.891)
(-1.567)
(-1.561)
(0.882)
(0.767)
(0.738)
(0.368)
Obj.
1sh
ort-term
elast.(size)
-0.00207
-0.00264
-0.00228
-0.00649
-0.00786
-0.00665
Obj.
1sh
ort-term
elast.(p
-valu
e)
0.353
0.337
0.417
0.0194
0.0169
0.0594
Obj.
1long-term
elast.(size)
0.00122
-0.00390
-0.00495
-0.00429
-0.00957
-0.0175
-0.0238
-0.0206
Obj.
1long-term
elast.(p
-valu
e)
0.657
0.358
0.347
0.433
0.370
0.00929
0.0115
0.0456
Obj.
2sh
ort-term
elast.(size)
-0.000936
-0.00254
-0.00209
0.00481
0.00628
0.00719
Obj.
2sh
ort-term
elast.(p
-valu
e)
0.654
0.232
0.390
0.419
0.176
0.164
Obj.
2long-term
elast.(size)
-0.00218
-0.00176
-0.00477
-0.00393
0.00241
0.0130
0.0190
0.0222
Obj.
2long-term
elast.(p
-valu
e)
0.545
0.545
0.201
0.366
0.842
0.435
0.194
0.186
Obj.
3sh
ort-term
elast.(size)
0.00126
0.00153
0.000987
0.00628
0.0135
0.0164
Obj.
3sh
ort-term
elast.(p
-valu
e)
0.614
0.600
0.764
0.610
0.152
0.0265
Obj.
3long-term
elast.(size)
0.00246
0.00237
0.00287
0.00185
0.0133
0.0169
0.0410
0.0506
Obj.
3long-term
elast.(p
-valu
e)
0.504
0.613
0.602
0.763
0.486
0.613
0.129
0.0172
Woold
ridgete
stAR(1
)(p
-valu
e)
00
00
0AR(1
)(p
-valu
e)
0.000676
0.000236
7.59e-0
55.49e-0
5AR(2
)(p
-valu
e)
0.238
0.217
0.208
0.226
Hanse
n(p
-valu
e)
0.858
0.466
0.255
0.340
No.ofin
stru
ments
57
81
105
129
No.ofobse
rvations
959
959
959
959
959
959
959
959
959
No.ofre
gions
129
129
129
129
129
129
129
129
129
Notes:In
colu
mns(1
)to
(5)standard
errors
are
calculate
daccord
ing
toNewey
and
West
(1987),
t-statisticsare
reported
inpare
nth
ese
s.In
colu
mns(6
)to
(9)z-sta
tistics
are
listed
inpare
nth
ese
sapplyin
gth
etw
o-ste
psy
stem
GM
Mestim
ato
ras
pro
pose
dby
(Blu
ndell
and
Bond,1998).
The
lagged
dependent
variable,com
pensa
tion
per
employee,low-skilled,mark
etpote
ntialand
thestru
ctu
ralfu
ndsvariablesare
assumed
tobeendogenous.
Wein
stru
mentth
eendogenousvariableswith
both
itslagsand
itsdiff
ere
nced
lagsand
use
the“collapse
”option.Sta
ndard
errors
are
correcte
dusing
theappro
ach
by
Win
dm
eijer(2
005).
*signifi
cantat10%
;**
signifi
cantat5%
;***
signifi
cantat1%.
34ECBWorking Paper Series No 1403December 2011
Table
7:Reduced-form
employmentmodel
excludingmarketpotential
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Obj.
X=
Obj.
1+2+3
Obj.
X=
Obj.
1
Emp.perwp.(t-1
)0.622***
0.606***
0.692***
0.756***
0.832***
0.639***
0.613***
0.670***
0.692***
0.704***
(5.997)
(8.186)
(10.10)
(10.31)
(14.90)
(6.133)
(9.324)
(12.25)
(14.60)
(15.67)
Comp.emp.(t-1
)0.0223
0.0347
-0.00402
-0.0313
-0.0255
0.0203
-0.00679
-0.0575
-0.0508
-0.0467
(0.476)
(0.740)
(-0.108)
(-1.046)
(-0.861)
(0.463)
(-0.202)
(-1.623)
(-1.846)
(-1.315)
Pop.young
(t-1
)-0
.0569
-0.0775*
-0.0461
-0.0497
-0.0317
-0.0695
-0.0689
-0.0283
-0.0250
-0.00977
(-1.482)
(-2.427)
(-1.479)
(-1.688)
(-1.494)
(-1.560)
(-1.941)
(-0.825)
(-1.114)
(-0.332)
Low-skilled
(t-1
)-0
.111*
-0.0953**
-0.130**
-0.0943*
-0.107*
-0.102*
-0.0894**
-0.117**
-0.0937*
-0.0707
(-2.317)
(-2.767)
(-2.610)
(-2.144)
(-2.569)
(-2.228)
(-2.751)
(-2.904)
(-2.207)
(-1.738)
Grr
(t-1
)0.00311
0.00269
0.00360
0.00149
0.00177
0.00228
0.00161
0.00105
-0.00261
-0.00202
(0.552)
(0.457)
(0.566)
(0.302)
(0.462)
(0.420)
(0.250)
(0.178)
(-0.523)
(-0.401)
Union
density
(t-1
)-0
.00139
-0.000934
0.000473
-0.000602
0.000793
-0.00183
-0.00238
-0.00185
-0.00158
-0.000769
(-0.349)
(-0.236)
(0.121)
(-0.142)
(0.252)
(-0.445)
(-0.625)
(-0.479)
(-0.401)
(-0.150)
SF
pcObj.
X(t-1
)-0
.00272
-0.00142
-0.00616
-0.00565
-0.00499
-0.00306
-0.00875
-0.00722
(-0.503)
(-0.296)
(-0.889)
(-0.873)
(-1.268)
(-0.901)
(-1.862)
(-1.686)
SF
pcObj.
X(t-2
)0.000630
-0.000688
-0.00105
-0.00182
-0.00102
-0.00164
(0.314)
(-0.207)
(-0.404)
(-1.036)
(-0.416)
(-0.759)
SF
pcObj.
X(t-3
)-0
.000274
-0.000463
0.00183
0.00126
(-0.129)
(-0.262)
(0.806)
(0.558)
SF
pcObj.
X(t-4
)0.00188
0.00138
(0.985)
(0.778)
SF
pcObj.
2(t-1
)0.00126
0.00418
0.00677
0.00601
(0.232)
(0.541)
(0.980)
(0.980)
SF
pcObj.
2(t-2
)0.000695
-0.000772
0.000554
(0.207)
(-0.203)
(0.173)
SF
pcObj.
2(t-3
)-0
.00129
-0.000418
(-0.563)
(-0.198)
SF
pcObj.
2(t-4
)0.00148
(0.953)
SF
pcObj.
3(t-1
)0.00796
0.0156*
0.0127*
0.0194*
(0.914)
(2.309)
(1.998)
(2.157)
SF
pcObj.
3(t-2
)-0
.00425
-0.000469
-0.00249
(-1.046)
(-0.104)
(-0.386)
SF
pcObj.
3(t-3
)0.00503
0.00445
(1.447)
(1.385)
SF
pcObj.
3(t-4
)-0
.00454
(-0.952)
Constant
0.00293
0.00261
0.00291
0.00207
0.00189
0.00254
0.00267
0.00333
0.00167
0.00129
(0.993)
(0.876)
(0.937)
(1.101)
(1.031)
(0.869)
(1.019)
(1.156)
(0.724)
(0.531)
Obj.
Xsh
ort-term
elast.(size)
-0.000786
-0.00712
-0.00528
-0.00488
-0.00795
-0.00623
Obj.
Xsh
ort-term
elast.(p
-valu
e)
0.860
0.361
0.380
0.120
0.0243
0.124
Obj.
Xlong-term
elast.(size)
-0.00690
-0.00255
-0.0292
-0.0314
-0.0129
-0.0148
-0.0258
-0.0210
Obj.
Xlong-term
elast.(p
-valu
e)
0.617
0.861
0.308
0.318
0.184
0.0778
0.0145
0.0870
Obj.
2sh
ort-term
elast.(size)
0.00488
0.00471
0.00763
Obj.
2sh
ort-term
elast.(p
-valu
e)
0.396
0.206
0.121
Obj.
2long-term
elast.(size)
0.00326
0.0148
0.0153
0.0257
Obj.
2long-term
elast.(p
-valu
e)
0.820
0.415
0.226
0.154
Obj.
3sh
ort-term
elast.(size)
0.0113
0.0173
0.0168
Obj.
3sh
ort-term
elast.(p
-valu
e)
0.123
0.0260
0.00881
Obj.
3long-term
elast.(size)
0.0205
0.0343
0.0561
0.0566
Obj.
3long-term
elast.(p
-valu
e)
0.349
0.121
0.0208
0.00314
AR(1
)(p
-valu
e)
0.00251
0.00139
0.000290
0.000288
0.000139
0.00207
0.000398
8.16e-0
56.07e-0
54.83e-0
5AR(2
)(p
-valu
e)
0.251
0.250
0.274
0.293
0.332
0.255
0.234
0.205
0.205
0.232
Hanse
n(p
-valu
e)
0.200
0.171
0.258
0.206
0.226
0.174
0.491
0.436
0.346
0.280
No.ofin
stru
ments
26
34
42
50
58
26
50
74
98
122
No.ofobse
rvations
964
964
964
964
964
964
964
964
964
964
No.ofre
gions
130
130
130
130
130
130
130
130
130
130
Notes:In
colu
mns(1
)to
(5)standard
errors
are
calculate
daccord
ing
toNewey
and
West
(1987),
t-statisticsare
reported
inpare
nth
ese
s.In
colu
mns(6
)to
(9)z-sta
tistics
are
listed
inpare
nth
ese
sapplyin
gth
etw
o-ste
psy
stem
GM
Mestim
ato
ras
pro
pose
dby
(Blu
ndell
and
Bond,1998).
The
lagged
dependent
variable,com
pensa
tion
per
employee,low-skilled,mark
etpote
ntialand
thestru
ctu
ralfu
ndsvariablesare
assumed
tobeendogenous.
Wein
stru
mentth
eendogenousvariableswith
both
itslagsand
itsdiff
ere
nced
lagsand
use
the“collapse
”option.Sta
ndard
errors
are
correcte
dusing
theappro
ach
by
Win
dm
eijer(2
005).
*signifi
cantat10%
;**
signifi
cantat5%
;***
signifi
cantat1%.
35ECB
Working Paper Series No 1403December 2011
Table
8:Spatialpan
elmodel:Objective1+
2+3
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
Reduced-form
employmentmodelin
clu
din
gmark
etpote
ntial
Reduced-form
em
ploym
entm
odelexclu
din
gm
ark
etpote
ntial
γ0.34154***
0.34157***
0.34173***
0.3381***
0.33821***
0.35114***
0.35093***
0.34989***
0.34537***
0.34563***
-11.8197
-11.8221
-11.8487
-11.7104
-11.6907
-12.2548
-12.2495
-12.2401
-12.0608
-12.0516
ρ-0
.22456***
-0.22542***
-0.2276***
-0.22404***
-0.22412***
-0.23212***
-0.23312***
-0.23459***
-0.23023***
-0.23042***
(-5.1052)
(-5.1214)
(-5.1788)
(-5.1011)
(-5.1005)
(-5.2817)
(-5.3028)
(-5.3482)
(-5.2515)
(-5.2543)
Emp.perw.p
.(t-1
)0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
-11.3023
-11.3033
-11.31
-11.3159
-11.316
-11.3032
-11.3049
-11.3123
-11.3185
-11.3186
Comp.emp.(t-1
)-0
.01255
-0.012424
-0.011004
-0.010827
-0.010841
-0.009644
-0.009546
-0.008308
-0.008318
-0.008371
(-0.73969)
(-0.73231)
(-0.64912)
(-0.63959)
(-0.64038)
(-0.56861)
(-0.56295)
(-0.49082)
(-0.49225)
(-0.49532)
Pop.young
(t-1
)-0
.18847***
-0.19002***
-0.19126***
-0.19172***
-0.19161***
-0.18125***
-0.18357***
-0.18582***
-0.18671***
-0.18643***
(-4.3752)
(-4.4003)
(-4.4349)
(-4.4511)
(-4.4445)
(-4.206)
(-4.2476)
(-4.3069)
(-4.3342)
(-4.3244)
Low-skilled
(t-1
)-0
.043752**
-0.043146**
-0.040268**
-0.039195**
-0.039277**
-0.048713**
-0.04775**
-0.043891**
-0.042463**
-0.042674**
(-2.0353)
(-2.0042)
(-1.8692)
(-1.8214)
(-1.8215)
(-2.2723)
(-2.2232)
(-2.0412)
(-1.9771)
(-1.9834)
Grr
(t-1
)0.0020002
0.0020406
0.0022363
0.0023245
0.0023232
0.0031164
0.0031432
0.0032223
0.0032439
0.0032355
-0.15657
-0.15975
-0.17536
-0.18255
-0.18245
-0.24349
-0.24564
-0.25239
-0.25452
-0.25387
Union
density
(t-1
)0.012178
0.01226
0.012716
0.013553
0.013549
-0.001901
0.0093358
0.010231
0.01132
0.01132
-0.42196
-0.42486
-0.44141
-0.47111
-0.47097
(-0.43053)
-0.32308
-0.35485
-0.3932
-0.39322
Mark
etpote
ntial(t-1
)0.15636**
0.15343**
0.13591**
0.12651**
0.12627**
-2.2736
-2.2232
-1.9558
-1.8181
-1.8115
SF
pcObj.
1+2+3
(t-1
)-0
.0009578
-0.00062332
-0.00056028
-0.00056063
-0.001911
-0.001944
-0.002004
-0.002008
(-0.50043)
(-0.3249)
(-0.29243)
(-0.29261)
(-0.43281)
(-0.44125)
(-0.45575)
(-0.45651)
SF
pcObj.
1+2+3
(t-2
)-0
.0034714**
-0.0032124**
-0.0032201**
-0.0039358**
-0.0036218**
-0.003641**
(-1.9207)
(-1.7742)
(-1.7739)
(-2.1926)
(-2.0127)
(-2.0192)
SF
pcObj.
1+2+3
(t-3
)-0
.0030335**
-0.0030405**
-0.0032725**
-0.0032906**
(-1.7652)
(-1.7652)
(-1.9069)
(-1.9136)
SF
pcObj.
1+2+3
(t-4
)9.59E-0
50.0002672
-0.059439
-0.16563
Obj.
1+2+3
short-term
elast.(size)
-0.0009578
-0.0040948
-0.0068063
-0.0067253
-0.001319
-0.004829
-0.007699
-0.007469
Obj.
1+2+3
short-term
elast.(p
-valu
e)
-0.50043
0.12457
0.029199
0.050505
-0.69004
0.067509
0.012494
0.02882
Obj.
1+2+3
long-term
elast.(size)
-0.0014547
-0.0062204
-0.010283
-0.010162
-0.002032
-0.007428
-0.011761
-0.011414
Obj.
1+2+3
long-term
elast.(p
-valu
e)
0.63715
0.045153
0.00093857
0.0010769
0.51541
0.018151
0.0001848
0.0002834
Sum
|γ|+
|ρ|+
|p|
0.94807
0.94896
0.95129
0.9441
0.94429
0.96523
0.96601
0.96645
0.95756
0.95802
No.ofre
gions
130
130
130
130
130
130
130
130
130
130
Notes:
The
spatialdynamic
panelestim
ato
ruse
sa
quasi-m
axim
um
likelihood
estim
ato
rapplyin
gth
eM
atlab
routine
sarpaneljihaiby
Yu,de
Jong,and
Lee
(2008).
t-statisticsare
reported
pare
nth
ese
s;*
signifi
cantat10%;**
signifi
cantat5%;***
signifi
cantat1%.
36ECBWorking Paper Series No 1403December 2011
Table
9:Spatialpan
elmodel:Objective1,
2,3
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
Reduced-form
employmentmodelin
clu
din
gmark
etpote
ntial
Reduced-form
em
ploym
entm
odelexclu
din
gm
ark
etpote
ntial
γ0.337***
0.33722***
0.33746***
0.34609***
0.34714***
0.34677***
0.34689***
0.35502***
(-11.6425)
(-11.6595)
(-11.7298)
(-12.0723)
(-12.0977)
(-12.0952)
(-12.1556)
(-12.501)
ρ-0
.23042***
-0.23468***
-0.24356***
-0.22012***
-0.23835***
-0.2417***
-0.25018***
-0.22565***
(-5.231)
(-5.309)
(-5.5465)
(-5.0069)
(-5.4153)
(-5.4708)
(-5.6974)
(-5.1329)
Emp.perw.p
.(t-1
)0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
0.38197***
(-11.3309)
(-11.3468)
(-11.4942)
(-11.5887)
(-11.3346)
(-11.351)
(-11.4966)
(-11.5918)
Comp.emp.(t-1
)-0
.010529
-0.009244
-0.008821
-0.010644
-0.007605
-0.006396
-0.005928
-0.008137
(-0.62117)
(-0.54578)
(-0.52601)
(-0.63909)
(-0.44871)
(-0.37782)
(-0.35361)
(-0.48886)
Pop.young
(t-1
)-0
.20562***
-0.20722***
-0.19739***
-0.19394***
-0.19823***
-0.20029***
-0.19039***
-0.18719***
(-4.7044)
(-4.7243)
(-4.5362)
(-4.476)
(-4.5319)
(-4.5635)
(-4.3723)
(-4.3201)
Low-skilled
(t-1
)-0
.043395**
-0.040953**
-0.046785**
-0.042372**
-0.048301**
-0.044929**
-0.050513**
-0.045677**
(-2.0201)
(-1.893)
(-2.1718)
(-1.9737)
(-2.2532)
(-2.0789)
(-2.3459)
(-2.1292)
Grr
(t-1
)0.0025073
0.0026931
0.0019783
0.002664
0.0036454
0.0037598
0.0030686
0.0036533
(-0.19674)
(-0.21176)
(-0.1571)
(-0.21317)
(-0.28549)
(-0.29511)
(-0.24323)
(-0.29191)
Union
density
(t-1
)0.010906
0.010813
0.010265
0.0067268
0.0078653
0.0081114
0.0075839
0.004229
(-0.37865)
(-0.37606)
(-0.36049)
(-0.23796)
(-0.27263)
(-0.28167)
(-0.2659)
(-0.14943)
Mark
etpote
ntial(t-1
)0.16151**
0.15407**
0.15611**
0.14119**
(-2.3461)
(-2.2311)
(-2.279)
(-2.0701)
SF
pcObj.
1(t-1
)0.0016257
0.0023661
0.0032246
0.0031731
0.0012348
0.0020178
0.002893
0.0028602
(-0.82966)
(-1.1745)
(-1.601)
(-1.5836)
(-0.63082)
(-1.0022)
(-1.4366)
(-1.4286)
SF
pcObj.
1(t-2
)-0
.00349**
-0.001338
-0.001822
-0.0038228**
-0.001641
-0.002153
(-1.8771)
(-0.69695)
(-0.93888)
(-2.0579)
(-0.85433)
(-1.111)
SF
pcObj.
1(t-3
)-0
.0037218**
-0.0049366***
-0.0039226**
-0.0051717***
(-1.9798)
(-2.5941)
(-2.0837)
(-2.7169)
SF
pcObj.
1(t-4
)0.002582
0.002871
(-1.3937)
(-1.5508)
SF
pcObj.
2(t-1
)-0
.001663
-0.002241
-0.003181
-0.002844
-0.001702
-0.002253
-0.003192
-0.002835
(-0.84348)
(-1.1037)
(-1.5575)
(-1.399)
(-0.86099)
(-1.1071)
(-1.5588)
(-1.3916)
SF
pcObj.
2(t-2
)3.46E-0
5-0
.001182
-0.00076
-0.000295
-0.001527
-0.001026
(-0.017492)
(-0.59811)
(-0.38199)
(-0.14937)
(-0.77268)
(-0.51567)
SF
pcObj.
2(t-3
)0.0068472***
0.0075695***
0.0067059***
0.0074734***
(-3.7631)
(-4.1526)
(-3.6786)
(-4.093)
SF
pcObj.
2(t-4
)-0
.0062745***
-0.006344***
(-3.5706)
(-3.6034)
SF
pcObj.
3(t-1
)-0
.0063922**
-0.0061241**
-0.005091
-0.0057013*
-0.0063799**
-0.0062015**
-0.005196
-0.0057941*
(-1.8847)
(-1.7881)
(-1.4952)
(-1.6708)
(-1.8761)
(-1.8064)
(-1.5224)
(-1.6947)
SF
pcObj.
3(t-2
)-0
.002905
-0.001993
-0.002052
-0.002671
-0.001803
-0.001868
(-0.88708)
(-0.61087)
(-0.62985)
(-0.81415)
(-0.55129)
(-0.57246)
SF
pcObj.
3(t-3
)-0
.002178
-0.002246
-0.001927
-0.002015
(-0.72102)
(-0.74204)
(-0.63677)
(-0.66488)
SF
pcObj.
3(t-4
)0.0020849
0.0022415
(-0.68831)
(-0.7387)
Obj.
1sh
ort-term
elast.(size)
-0.001124
-0.001835
-0.001004
-0.001805
-0.00267
-0.001594
Obj.
1sh
ort-term
elast.(p
-valu
e)
0.67026
0.53127
0.74863
0.49187
0.35935
0.60994
Obj.
1long-term
elast.(size)
0.002452
-0.001696
-0.00277
-0.001535
0.0018914
-0.002763
-0.004088
-0.002471
Obj.
1long-term
elast.(p
-valu
e)
0.43369
0.59932
0.39138
0.63704
0.55187
0.39879
0.21287
0.45364
Obj.
2sh
ort-term
elast.(size)
-0.002207
0.002484
-0.002309
-0.002548
0.0019875
-0.002732
Obj.
2sh
ort-term
elast.(p
-valu
e)
0.41144
0.43535
0.52557
0.34311
0.53248
0.45286
Obj.
2long-term
elast.(size)
-0.002509
-0.003329
0.0037492
-0.003532
-0.002608
-0.003901
0.0030432
-0.004235
Obj.
2long-term
elast.(p
-valu
e)
0.42563
0.30485
0.2538
0.283
0.41606
0.2371
0.36188
0.20496
Obj.
3sh
ort-term
elast.(size)
-0.009029
-0.009262
-0.007915
-0.008873
-0.008925
-0.007436
Obj.
3sh
ort-term
elast.(p
-valu
e)
0.059633
0.089789
0.1742
0.064811
0.1028
0.20225
Obj.
3long-term
elast.(size)
-0.009641
-0.013623
-0.013979
-0.012104
-0.009772
-0.013583
-0.013666
-0.011529
Obj.
3long-term
elast.(p
-valu
e)
0.076133
0.013337
0.010695
0.029246
0.077507
0.015234
0.014076
0.040788
Sum
|γ|+
|ρ|+
|p|
0.94938
0.95387
0.96298
0.94818
0.96746
0.97044
0.97904
0.96263
No.ofobse
rvations
No.ofre
gions
130
130
130
130
130
130
130
130
Notes:
The
spatialdynamic
panelestim
ato
ruse
sa
quasi-m
axim
um
likelihood
estim
ato
rapplyin
gth
eM
atlab
routine
sarpaneljihaiby
Yu,de
Jong,and
Lee
(2008).
t-statisticsare
reported
pare
nth
ese
s;*
signifi
cantat10%;**
signifi
cantat5%;***
signifi
cantat1%.
37ECB
Working Paper Series No 1403December 2011
Table
10:Sizeof
theestimated
spatialcoeffi
cients
fordifferentweigh
tmatrices(W
)
Obj.
1,2,3
sfnotin
clu
ded
sfup
to1
lag
sfup
to2
lags
sfup
to3
lags
sfup
to4
lags
Wλ
ργ
λρ
γλ
ργ
λρ
γλ
ργ
W2
0.32856
-0.1865
0.38197
0.32413
-0.18943
0.38197
0.32482
-0.19205
0.38197
0.32448
-0.19767
0.38197
0.33392
-0.17771
0.38197
W3
0.34154
-0.22456
0.38197
0.337
-0.23042
0.38197
0.33722
-0.23468
0.38197
0.33746
-0.24356
0.38197
0.34609
-0.22012
0.38197
W4
0.37156
-0.34118
0.53699
0.36791
-0.34606
0.54398
0.36716
-0.34877
0.53999
0.36603
-0.35288
0.52999
0.37292
-0.32835
0.51296
W5
0.38225
-0.38921
0.56596
0.37802
-0.39406
0.57
0.37788
-0.39841
0.56397
0.37713
-0.40448
0.55296
0.38329
-0.38017
0.53499
W6
0.38957
-0.45783
0.60199
0.38465
-0.46014
0.59195
0.385
-0.46692
0.58998
0.3847
-0.47585
0.57598
0.3915
-0.448
0.56698
W7
0.38546
-0.49458
0.602
0.38096
-0.49693
0.60196
0.38001
-0.50066
0.585
0.38017
-0.51256
0.575
0.38776
-0.48332
0.567
W8
0.37296
-0.50608
0.59696
0.36857
-0.50748
0.59099
0.36812
-0.51292
0.591
0.36687
-0.52166
0.56696
0.37688
-0.49287
0.55698
W9
0.36686
-0.51819
0.59398
0.36322
-0.52285
0.60295
0.36245
-0.52942
0.59895
0.36123
-0.53723
0.57899
0.37179
-0.50432
0.576
W10
0.36616
-0.56535
0.58698
0.36247
-0.57089
0.59299
0.36141
-0.57675
0.59097
0.35926
-0.58485
0.55697
0.37104
-0.55318
0.56397
W11
0.3632
-0.5758
0.589
0.35949
-0.58131
0.59398
0.35832
-0.58506
0.589
0.35657
-0.59123
0.56996
0.36768
-0.55659
0.557
W12
0.36044
-0.58417
0.60098
0.356
-0.59104
0.60599
0.35409
-0.5927
0.587
0.35255
-0.60396
0.568
0.36373
-0.56673
0.56499
W13
0.35468
-0.57504
0.61299
0.34913
-0.58129
0.60198
0.34858
-0.58531
0.60397
0.34682
-0.59689
0.583
0.35775
-0.55863
0.57
W14
0.34763
-0.58281
0.60298
0.34214
-0.58981
0.59196
0.34156
-0.59462
0.59594
0.33854
-0.59996
0.56499
0.35126
-0.56369
0.54796
W15
0.34088
-0.56878
0.61199
0.33552
-0.57599
0.606
0.33453
-0.57942
0.59497
0.33263
-0.58849
0.585
0.34595
-0.54928
0.57199
W.d
ist
0.34763
-0.58281
0.60298
0.34214
-0.58981
0.59196
0.34156
-0.59462
0.59594
0.33854
-0.59996
0.56499
0.35126
-0.56369
0.54796
W.d
ist2
0.37963
-0.67363
0.71899
0.37576
-0.6856
0.70098
0.37627
-0.69549
0.706
0.37476
-0.70408
0.68798
0.38238
-0.66052
0.671
Obj.
1+2+3
sfnotin
clu
ded
sfup
to1
lag
sfup
to2
lags
sfup
to3
lags
sfup
to4
lags
Wλ
ργ
λρ
γλ
ργ
λρ
γλ
ργ
W2
0.32856
-0.1865
0.38197
0.32853
-0.18699
0.38197
0.32833
-0.1879
0.38197
0.32444
-0.18469
0.38197
0.32447
-0.1847
0.38197
W3
0.34154
-0.22456
0.38197
0.34157
-0.22542
0.38197
0.34173
-0.2276
0.38197
0.3381
-0.22404
0.38197
0.33821
-0.22412
0.38197
W4
0.37156
-0.34118
0.53699
0.37175
-0.34237
0.54299
0.37158
-0.3437
0.54497
0.36842
-0.33965
0.54999
0.36809
-0.33899
0.54397
W5
0.38225
-0.38921
0.56596
0.38259
-0.3913
0.57298
0.38222
-0.39189
0.57397
0.37909
-0.38781
0.57195
0.37933
-0.38805
0.57399
W6
0.38957
-0.45783
0.60199
0.38936
-0.45883
0.59698
0.38917
-0.46041
0.59796
0.3866
-0.45642
0.60199
0.38631
-0.45616
0.60096
W7
0.38546
-0.49458
0.602
0.38511
-0.49515
0.59599
0.38539
-0.49798
0.60598
0.38233
-0.49289
0.59797
0.38259
-0.49353
0.61098
W8
0.37296
-0.50608
0.59696
0.37285
-0.50681
0.59498
0.3734
-0.51314
0.59497
0.37147
-0.51093
0.60598
0.37055
-0.51004
0.59398
W9
0.36686
-0.51819
0.59398
0.36725
-0.51999
0.60498
0.36813
-0.52801
0.61095
0.36588
-0.52541
0.611
0.36531
-0.52526
0.611
W10
0.36616
-0.56535
0.58698
0.36669
-0.56725
0.60195
0.36705
-0.57349
0.60299
0.36466
-0.5707
0.59599
0.36406
-0.57073
0.596
W11
0.3632
-0.5758
0.589
0.36329
-0.57664
0.594
0.36388
-0.58273
0.60396
0.36158
-0.58031
0.598
0.36079
-0.58046
0.59696
W12
0.36044
-0.58417
0.60098
0.36034
-0.58487
0.603
0.3603
-0.58825
0.60498
0.35834
-0.5859
0.61096
0.35717
-0.58539
0.60297
W13
0.35468
-0.57504
0.61299
0.35454
-0.57547
0.61295
0.35433
-0.57913
0.60799
0.35195
-0.57747
0.60296
0.35107
-0.57762
0.59997
W14
0.34763
-0.58281
0.60298
0.34718
-0.58234
0.59499
0.34737
-0.58815
0.598
0.34503
-0.58702
0.59497
0.3442
-0.58805
0.59497
W15
0.34088
-0.56878
0.61199
0.34054
-0.56834
0.60597
0.34082
-0.57492
0.611
0.33852
-0.57399
0.60997
0.33768
-0.57473
0.60894
W.d
ist
0.34763
-0.58281
0.60298
0.34718
-0.58234
0.59499
0.34737
-0.58815
0.598
0.34503
-0.58702
0.59497
0.3442
-0.58805
0.59497
W.d
ist2
0.37963
-0.67363
0.71899
0.37934
-0.67385
0.71098
0.37944
-0.67817
0.70899
0.37699
-0.67499
0.72197
0.37616
-0.67349
0.70497
Notes:
The
spatialdynamic
panelestim
ato
ruse
sa
quasi-m
axim
um
likelihood
estim
ato
rapplyin
gth
eM
atlab
routine
sarpaneljihaiby
Yu,de
Jong,and
Lee
(2008).
Irre
spective
ofwhich
weightmatrix
isuse
d,all
indicato
rsare
statistically
signifi
cantatth
e1%
level.
The
coeffi
cients
referto
equation
(2)and
corresp
ond
toW
em
pi,t
(λ),
Wem
pi,t
−1
(ρ)and
em
pi,t
−1
(γ).
38ECBWorking Paper Series No 1403December 2011
Table 11: Interaction model: Objectives 1+2+3
(1) (2) (3) (4)
Emp. per wp. (t-1) 0.611*** 0.588*** 0.545*** 0.727***(6.164) (5.010) (7.929) (9.349)
Comp. emp. (t-1) 0.0417 0.0722 0.0967** 0.103(1.083) (1.450) (2.880) (1.893)
Pop. young (t-1) -0.0626 -0.0670 -0.0912* -0.0430(-1.452) (-1.444) (-2.526) (-0.988)
Grr (t-1) 0.00407 0.00219 0.00608 0.00190(0.691) (0.478) (0.870) (0.236)
Union density (t-1) -0.00178 -0.000570 -0.00189 -0.00253(-0.489) (-0.131) (-0.500) (-0.551)
SF pc Obj. 1+2+3 (t-1) -0.00325 -0.00137 -0.000596 -0.00301(-0.687) (-0.283) (-0.140) (-0.670)
SF pc Obj. 1+2+3 x Low-skilled (t-1) -0.0684 -0.0562 -0.0213 -0.0411(-1.507) (-1.342) (-0.437) (-0.686)
Low-skilled (t-1) 0.156 0.189 0.200 0.347(1.161) (1.303) (1.292) (1.054)
SF pc Obj. 1+2+3 (t-2) 0.000102 -0.00113 -0.000849(0.0476) (-0.462) (-0.232)
SF pc Obj. 1+2+3 x Low-skilled (t-2) -0.0116 -0.0219 -0.00415(-0.458) (-0.640) (-0.0726)
Low-skilled (t-2) -0.0587 -0.108 -0.358(-0.621) (-0.887) (-1.187)
SF pc Obj. 1+2+3 (t-3) 0.000281 -0.000613(0.117) (-0.235)
SF pc Obj. 1+2+3 x Low-skilled (t-3) -0.00939 0.0205(-0.684) (0.988)
Low-skilled (t-3) 0.0316 -0.0485(0.588) (-0.568)
SF pc Obj. 1+2+3 (t-4) -0.00125(-0.575)
SF pc Obj. 1+2+3 x Low-skilled (t-4) 0.00517(0.287)
Low-skilled (t-4) -0.0269(-0.366)
Constant 0.0142 0.00712 0.00649 0.0221(0.903) (0.475) (0.357) (1.207)
Obj. 1+2+3 short-term elast. (size) -0.00127 -0.00144 -0.00572Obj. 1+2+3 short-term elast. (p-value) 0.788 0.784 0.294Obj. 1+2+3 long-term elast. (size) -0.00835 -0.00308 -0.00317 -0.0209Obj. 1+2+3 long-term elast. (p-value) 0.510 0.783 0.784 0.266AR(1) (p-value) 0.00115 0.00961 0.0189 0.00324AR(2) (p-value) 0.246 0.606 0.186 0.411Hansen (p-value) 0.606 0.560 0.358 0.200No. of instruments 40 43 46 49No. of observations 964 834 705 576No. of regions 130 130 129 129
Notes: z-statistics are listed in parentheses applying the two-step system GMM estimator as proposed by (Blundelland Bond, 1998). The lagged dependent variable, compensation per employee, low-skilled, market potential and thestructural funds variables are assumed to be endogenous. We instrument the endogenous variables with both its lags andits differenced lags and use the “collapse” option. Standard errors are corrected using the approach by Windmeijer (2005).* significant at 10%; ** significant at 5%; *** significant at 1%.
Figure 2: Marginal effects of structural funds on employment
Obj. 1+2+3
short-term
−.0
4−
.02
0.0
2.0
4M
argi
nal e
ffect
s of
L1.
Obj
. 123
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.0
50
.05
Mar
gina
l effe
cts
of L
1.O
bj. 1
23
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 1
short-term
−.0
4−
.02
0.0
2.0
4M
argi
nal e
ffect
s of
Obj
. 1_1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.0
6−
.04
−.0
20
.02
.04
Mar
gina
l effe
cts
of L
1.O
bj. 1
−2−
3_1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L2.
Obj
. 123
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L2.
Obj
. 123
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
4−
.02
0.0
2.0
4M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
6−
.04
−.0
20
.02
.04
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 123
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 123
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
4−
.02
0.0
2.0
4.0
6M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
1
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 2
short-term
−.0
2−
.01
0.0
1.0
2M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.0
2−
.01
0.0
1.0
2.0
3M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 3
short-term
−.0
4−
.02
0.0
2.0
4.0
6M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term−
.05
0.0
5.1
Mar
gina
l effe
cts
of L
1.O
bj. 1
−2−
3_3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
4−
.02
0.0
2.0
4M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
4−
.02
0.0
2.0
4M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
20
.02
.04
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
0.0
1.0
2.0
3.0
4.0
5M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
Mar
gina
l effe
cts
of L
3.O
bj. 1
−2−
3_2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
20
.02
.04
.06
Mar
gina
l effe
cts
of L
3.O
bj. 1
−2−
3_2
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
.02
.04
.06
.08
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
3
−.5 0 .5Low skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Notes: The estimation results are based on the baseline specification Reduced-form employment model including marketpotential displayed in equation (3). The regressions are estimated using the two-step system GMM estimator proposedby Blundell and Bond (1998), while standard errors are corrected using the approach by Windmeijer (2005). The laggeddependent variable, compensation per employee, low-skilled, market potential and the structural funds variables areassumed to be endogenous. We instrument the endogenous variables with both its lags and its differenced lags and usethe “collapse” option. The marginal effects are calculated for short-term and long-term elasticities as well as for one toup to three lags.
39ECB
Working Paper Series No 1403December 2011
Figure 3: Marginal effects of structural funds on employment (Reduced-form employment model including market potential using thehigh skilled)
Obj. 1+2+3
short-term
−.1
−.0
50
.05
Mar
gina
l effe
cts
of L
1.O
bj. 1
23
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L1.
Obj
. 123
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 1
short-term
−.0
4−
.02
0.0
2M
argi
nal e
ffect
s of
Obj
. 1_1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.0
5−
.04
−.0
3−
.02
−.0
10
Mar
gina
l effe
cts
of L
1.O
bj. 1
−2−
3_1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
Mar
gina
l effe
cts
of L
2.O
bj. 1
23
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
5−
.1−
.05
0.0
5.1
Mar
gina
l effe
cts
of L
2.O
bj. 1
23
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
6−
.04
−.0
20
.02
.04
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 123
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.2
−.1
0.1
Mar
gina
l effe
cts
of L
3.O
bj. 1
23
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
Mar
gina
l effe
cts
of L
3.O
bj. 1
−2−
3_1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
1
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 2
short-term
−.0
6−
.04
−.0
20
.02
.04
Mar
gina
l effe
cts
of L
1.O
bj. 1
−2−
3_2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Obj. 3
short-term
−.0
50
.05
.1M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
long-term
−.0
50
.05
.1.1
5M
argi
nal e
ffect
s of
L1.
Obj
. 1−
2−3_
3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
5−
.1−
.05
0.0
5.1
Mar
gina
l effe
cts
of L
2.O
bj. 1
−2−
3_2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1.1
5M
argi
nal e
ffect
s of
L2.
Obj
. 1−
2−3_
3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
Mar
gina
l effe
cts
of L
3.O
bj. 1
−2−
3_2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.1
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
2
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
−.0
50
.05
.1M
argi
nal e
ffect
s of
L3.
Obj
. 1−
2−3_
3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
0.0
5.1
.15
Mar
gina
l effe
cts
of L
3.O
bj. 1
−2−
3_3
−1 −.5 0 .5 1High skilled
Dashed lines correspond to lower and upper confidence interval bounds.
Notes: The estimation results are based on the baseline specification of the reduced-form employment model includingmarket potential and interacting the structural funds variable with the share of high-skilled population. The regressionsare estimated using the two-step system GMM estimator proposed by Blundell and Bond (1998), while standard errorsare corrected using the approach by Windmeijer (2005). The lagged dependent variable, compensation per employee,high-skilled, market potential and the structural funds variables are assumed to be endogenous. We instrument theendogenous variables with both its lags and its differenced lags and use the “collapse” option. The marginal effects arecalculated for short-term and long-term elasticities as well as for one to up to three lags.34
40ECBWorking Paper Series No 1403December 2011
41ECB
Working Paper Series No 1403December 2011
C Calculation of the interaction effects
We estimate an interaction model, interacting two variables, namely struc-tural funds (sf) and percentage share of low-skilled population (z). Themarginal effects are calculated by taking the first derivative of our specifica-tion listed in equation (3), i.e.:
∂ emp
∂ sf= βL1.sf + βL1.sf ·z · z
where L. denotes the use of a lagged variable. The level of uncertainty regard-ing the marginal effects is indicated by the variance (V ar) of the marginaleffects. If the marginal effects consists of two addends (as it is the case in theequation above), the variance of the short-term elasticity can be calculatedas follows:
V ar
(∂ emp.
∂ sf
)=V ar(βL1.sf ) + z2 V ar(βL1.sf ·z)+
+2 z Cov(βL1.sf , βL1.z)
Generally, if the marginal effects consists of more than two addends, thevariance can be approximated using the following Taylor rule,
V ar (g(X, Y )) ∼(∂g(X, Y )
∂X
)2
· V ar(X) +
(∂g(X,Y )
∂Y
)2
· V ar(Y )+
+ 2
(∂g(X, Y )
∂X· ∂g(X, Y )
∂Y· Cov(X,Y )
)where g(X,Y ) stands for the function of the marginal effects.This implies that the long-term elasticity is calculated as:
V ar
(∂ emp.
∂ sf
)=(V ar(βL1.sf ) + z2 V ar(βL1.sf ·z)+
+2 z Cov(βL1.sf , βL1.z)) · (1− βL1.emp)−1
If the estimation equation includes the structural funds variable with up totwo lags, the marginal effects are computed via the following expression:
∂ emp
∂ sf= βL1.sf + βL2.sf + z (βL1.sf ·z + βL2.sf ·z)
The variance of the short-term elasticity is then defined as:
42ECBWorking Paper Series No 1403December 2011
V ar
(∂emp.
∂sf
)= V ar(βL1.sf ) + V ar(βL2.sf ) + z2 V ar(βL1.sf ·z)+
+ z2 V ar(βL2.sf ·z) + 2Cov(βL1.sf , βL2.sf )+
+ 2 z Cov(βL1.sf , βL1.sf ·z) + 2 z Cov(βL1.sf , βL2.sf ·z)+
+ 2 z Cov(βL2.sf , βL1.sf ·z) + 2 z Cov(βL2.sf , βL2.sf ·z)+
+ 2 z2Cov(βL1.sf ·z, βL2.sf ·z)
whereas the variance of the dynamic long-term elasticity is given by:
V ar
(∂emp.
∂sf
)= (V ar(βL1.sf ) + V ar(βL2.sf ) + z2 V ar(βL1.sf ·z)+
+ z2 V ar(βL2.sf ·z) + 2Cov(βL1.sf , βL2.sf )+
+ 2 z Cov(βL1.sf , βL1.sf ·z) + 2 z Cov(βL1.sf , βL2.sf ·z)+
+ 2 z Cov(βL2.sf , βL1.sf ·z) + 2 z Cov(βL2.sf , βL2.sf ·z)+
+ 2 z2Cov(βL1.sf ·z, βL2.sf ·z)) · (1− βL1.emp)−1
Finally, we take account of the lower and upper bound of the 95% confidenceintervals, which can be calculated as follows:
∂ emp
∂ sf± tdf,p
√V ar(d emp/d sf),
using the inverse t-distribution function to create the multiplier. tdf,p is thecritical value in a t-distribution and df stands for the degrees of freedom(n − k), where n refers to the number of observations and k refers to thenumber or regressors, including the intercept, that produces a p-value atwhich hypothesis tests are to be made.
43ECB
Working Paper Series No 1403December 2011
ReferencesBasile, R., and L. de Benedicits (2008): “Regional unemployment and productivity
in Europe,” Papers in Regional Science, 87(2), 182.
Becker, S., P. Egger, and M. von Ehrlich (2010): “Going NUTS: The effect ofEU Structural Funds on regional performance,” Journal of Public Economics, 94(9-10),578–590.
Berman, E., L. Bound, and Z. Griliches (1994): “Changes in the demand for skilledlabor within U.S. manufacturing: Evidence from the annual survey of manufactures,”The Quarterly Journal of Economics, 109(2), 367–397.
Blau, F., and L. Kahn (1999): “Institutions and laws in the labor market,” in Handbookof Labor Economics, ed. by O. Ashenfelter, and D. Card, vol. 3A, pp. 1399–1461. North-Holland, Amsterdam.
Blundell, R., and S. Bond (1998): “Initial conditions and moment restrictions indynamic panel data models,” Journal of Econometrics, 87, 115–143.
Bond, S., A. Hoeffler, and J. Temple (2001): “GMM estimation of empirical growthmodels,” Nuffield College Economics Working Paper, 2001-W21.
Bondonio, D., and R. Greenbaum (2006): “Do business investment incentives pro-mote employment in declining areas? Evidence from EU Objective-2 regions,” EuropeanUrban and Regional Studies, 13, 225–244.
Bouvet, F. (2005): “European Union regional policy: allocation determinants and effectson regional economic growth,” Working Paper, Department of Economics, Universityof California, Davis.
Bowsher, C. (2002): “On testing overidentifying restrictions in dynamic panel datamodels,” Economic Letters, 77, 211–220.
Brambor, T., W. Clark, and M. Golder (2006): “Understanding interaction models:Improving empirical analyses,” Political Analysis, 14, 63–82.
Braumoeller, B. (2004): “Hypothesis testing and multiplicative interaction terms,”International Organization, 58, 807–820.
Busch, B., K. Lichtblau, and C. Schnabel (1998): “Kohasionspolitik, Konvergenzund Arbeitslosigkeit in der Europaischen Union: Eine empirische Analyse mit Region-aldaten,” Jahrbuch fur Wirtschaftswissenschaften, 49, 1–25.
Dall’erba, S., and J. Le Gallo (2007): “The impact of EU regional support ongrowth and employment,” Czech Journal of Economics and Finance, 57(7), 325–340.
(2008): “Regional convergence and the impact of European structural funds 1989-1999: A spatial econometric analysis,” Papers in Regional Science, 82(2), 219–244.
Driscoll, J., and A. Kraay (1998): “Consistent covariance matrix estimation withspatially dependent panel data,” The Review of Economics and Statistics, 80(4), 549–560.
44ECBWorking Paper Series No 1403December 2011
Elhorst, P. (2003): “The mystery of regional unemployment differentials: A survey oftheoretical and empirical explanations,” Journal of Economic Surveys, 17, 709–748.
Elhorst, P., and A. Zeilstra (2007): “Labour force participation rates at the regionaland national levels of the European Union: An integrated analysis,” Papers in RegionalScience, 86(4), 525–549.
Elhorst, P. J. (2010): “Spatial panel data models,” in Handbook of applied spatialanalysis, ed. by M. Fischer, and A. Getis, pp. 377–407. Springer, Berlin.
Ertur, C., and W. Koch (2006): “Regional disparities in the European Union andthe enlargement process: An exploratory spatial data analysis, 1995-2000,” Journal ofRegional Science, 40, 723–765.
Esposti, R., and S. Bussoletti (2008): “Impact of Objective 1 funds on regionalgrowth convergence in the European Union: A panel-data approach,” Regional Studies,42(2), 159–173.
European Commission (1995): The implementation of the reform of the structural funds1993: Fifth annual report. European Commission, Brussels.
(1996a): Seventh annual report on the structural funds 1995. European Commis-sion, Brussels.
(1996b): Sixth annual report on the structural funds 1994. European Commission,Brussels.
(1997): Eighth annual report on the structural funds 1996. European Commission,Brussels.
(1998): Ninth annual report on the structural funds 1997. European Commission,Brussels.
(1999): Tenth annual report on the structural funds 1998. European Commission,Brussels.
(2000): Report from the Commission. 11th annual report on the structural funds1999. European Commission, Brussels.
(2007): “Commission regulation (EC) No. 105/2007,” Official Journal of theEuropean Union, L 39.
Fitzenberger, B., K. Kohn, and Q. Wang (2011): “The erosion of union mem-bership in Germany: determinants, densities, decompositions,” Journal of PopulationEconomics, 24(1), 141–165.
Griliches, Z. (1969): “Capital-skill complementarity,” The review of Economics andStatistics, 51(4), 465–468.
Hagen, T., and P. Mohl (2011a): “Does EU Cohesion policy really increase publicinvestment?,” ECB Working Paper (forthcoming).
45ECB
Working Paper Series No 1403December 2011
(2011b): “Econometric evaluation of EU Cohesion Policy: A survey,” in Inter-national Handbook of Economic Integration, ed. by M. Jovanovic, vol. 3, pp. 343–370.Edward Elgar, Cheltenham (UK) and Northampton (USA).
Heinemann, F., P. Mohl, and S. Osterloh (2009): “Who’s afraid of an EU taxand why? Revenue system preferences in the European Parliament,” The Review ofInternational Organizations, 4(1), 73–99.
Holmlund, B. (1998): “Unemployment insurance in theory and practice,” ScandinavianJournal of Economics, 100(1), 113–141.
Kamps, C., N. Leiner-Killinger, and R. Martin (2009): “The cyclical impact of EUCohesion policy in fast growing countries,” Intereconomics, January/February, 23–29.
Lazear, E. (1990): “Job security provisions and employment,” Quarterly Journal ofEconomics, 105, 699–726.
Le Gallo, J., and C. Ertur (2003): “Exploratory spatial data analysis of the distri-bution of regional per capita GDP in Europe, 1980-1995,” Papers in Regional Science,82, 175–201.
LeSage, J., and M. Fischer (2008): “Spatial growth regressions: Model specification,estimation and interpretation,” Spatial Economic Analysis, 3(3), 275–304.
LeSage, J., and K. R. Pace (2010): “The biggest myth in spatial econometrics,”mimeo, (version from 1 December 2010).
LeSage, J. P., and K. R. Pace (2004): Arc Mat, a toolbox for using ArcView shapefiles for spatial econometrics and statistics, vol. 3234/2004 of Lecture Notes in ComputerScience. Springer, Berlin, Heidelberg.
Mehrhoff, J. (2009): “A solution to the problem of too many instruments in dynamicpanel data GMM,” Deutsche Bundesbank Discussion Paper, 31/2009.
Mohl, P., and T. Hagen (2010): “Do EU structural funds promote regional growth?New evidence from various panel data approaches,” Regional Science and Urban Eco-nomics, 40, 353–365.
Newey, W., and K. West (1987): “A simple, positive semi-definite, heteroskedasticityand autocorrelation consistent covariance matrix,” Econometrica, 55(3), 703–708.
Nickell, S. (1987): “Dynamic models of labour demand, handbook of labor economics,”in Handbook of Labor Economics, ed. by O. Ashenfelter, and R. Layard, pp. 473–522.Elsevier, Amsterdam.
Nickell, S., and R. Layard (1999): “Labor market institutions and economic perfor-mance,” in Handbook of labor economics, ed. by O. Ashenfelter, and D. Card, vol. 3,pp. 3029–3084. North Holland, Amsterdam.
Nickell, S., L. Nunziata, and W. Ochel (2005): “Unemployment in the OECD sincethe 1960s. What do we know?,” The Economic Journal, 115(January), 1–27.
Pace, K. R. (2003): “Spatial Statistics Toolbox 2.0,” mimeo.
46ECBWorking Paper Series No 1403December 2011
Roodman, D. (2009): “A note on the theme of too many instruments,” Oxford Bulletinof Economics and Statistics, 71(1), 135–158.
Scarpetta, S. (1996): “Assessing the role of labour market policies and institutionalsettings on unemployment: a cross country study,” OECD Economic Studies, 26, 43–98.
Topel, R. (1986): “Local labor markets,” Journal of Political Economy, 94(3), 111–143.
Windmeijer, F. (2005): “A finite sample correction for the variance of linear efficienttwo-step GMM estimators,” Journal of Econometrics, 126(1), 25–51.
Wooldridge, J. M. (2002): Econometric analysis of cross section and panel data. TheMIT Press, Cambridge, MA.
Yu, J., R. de Jong, and L.-f. Lee (2008): “Quasi-maximum likelihood estimators forspatial dynamic panel data with fixed effects when both n and T are large,” Journal ofEconometrics, 146, 118–134.